DEV Community: dev game

Why Death Stranding 2 Bet on Decima: The Borrowed Dutch Engine Behind Kojima's Hiking Sequel

dev game — Thu, 22 Oct 2026 08:42:25 +0000

Death Stranding 2: On the Beach runs on Decima, not Unreal Engine 5 and not Fox Engine. That is the short answer to what engine Death Stranding 2 uses: Kojima Productions stayed on Guerrilla Games' proprietary Decima for the PlayStation 5 release on June 26, 2025 and the Windows port dated March 19, 2026.

That choice looks stubborn in 2025. Almost every open-world sequel at this scale rents Unreal. Kojima had every reason to switch: a bigger map across Mexico and Australia, more vehicles, more combat, and PlayStation 5 Pro Enhanced ambitions. He did not. My reading, separated from the facts below, is simple: Decima is a hiking engine, not a shooter engine, and Death Stranding 2 is still a hiking game wearing combat armor. Kojima picked the only AAA engine whose core fantasy is walking beautifully while other players quietly build the road for you.

What Decima actually is

Decima is a proprietary engine made by Guerrilla Games and released in November 2013, with tools for artificial intelligence and game physics plus support for 4K resolution and high-dynamic-range imaging. It now lists PlayStation 4, PlayStation 5, Windows, macOS, iOS, iPadOS and Xbox Series X/S as operating targets.

The credit list matters because it explains what the engine was tuned for:

Guerrilla's own note about the partnership is unusually blunt for engine marketing. The studio says it is sharing its engine with Kojima Productions and collaborating on the future of it, and that it set up a small Kojima Amsterdam outpost to make that work. Engine sharing here meant people sharing, not just code sharing.

Even the name is a loan story. Decima was originally just called the engine inside Guerrilla, then named for marketing in late 2016 to symbolize the Dutch-Japanese partnership after Dejima, the former Japanese island that hosted a Dutch trading post. A trading-post island is a good metaphor for what Death Stranding does: isolated nodes, foreign tools, slow connection.

Why Kojima could not just reuse Fox Engine

New players often ask whether Death Stranding uses Fox Engine because Kojima built Fox at Konami. The confusion is understandable, but the ownership answer is clean.

Fox Engine is a discontinued proprietary engine by Konami, started by Hideo Kojima after Metal Gear Solid 4 with the goal of making the best engine in the world. Its first commercial release was Pro Evolution Soccer 2014. After seven years on Pro Evolution Soccer, Konami discontinued Fox Engine in favor of Unreal Engine for PlayStation 5 and Xbox Series X development.

Death Stranding was the first game from Kojima Productions after its split from Konami in 2015. Kojima left the Fox Engine behind because he left the company that owned it. That homelessness explains the next step.

In June 2016 Kojima said he was inspecting two engine candidates, of which the latter had been used to create the first teaser shown at E3 2016. After receiving the Industry Icon award at The Game Awards 2016 he premiered a trailer with the engine logo, and at PlayStation Experience in December 2016 he announced the partnership with Guerrilla Games to use Decima for Death Stranding.

This is my interpretation, not a sourced quote: most studios pick an engine for rendering or hiring. Kojima picked Decima for shelter. He needed a working open-world editor, a PS4-ready renderer, and a team willing to hand over source in months, not years. Guerrilla offered all three plus an Amsterdam desk.

What Death Stranding 2 asks the engine to do

Death Stranding 2 is a third-person action-adventure game with asynchronous online functions, set primarily in Mexico and Australia as large open worlds. The core loop is still delivery: navigate difficult terrain while balancing cargo weight on Sam's backpack, then connect preppers to the Chiral Network.

The sequel adds more systems that stress terrain and infrastructure rather than pure shooting:

This list is why Unreal would have been the lazy answer and Decima was the specific answer. Unreal can do all of this. But Decima had already shipped two things Kojima needed: Horizon-style long-view landscapes where empty ground still looks meaningful, and a production-proven way to stream huge natural terrain on PlayStation without turning every rock into a loading screen.

What Decima gained Kojima, and what it cost

Gained: empty space that holds attention.

Horizon's reputation was built on vistas. Death Stranding reuses that strength for a different emotion. Where Horizon makes you scan for machines, Death Stranding makes you scan for slope, river depth, and Timefall damage. Decima's 4K and HDR pipeline turns a 10-minute walk into visual content. That sounds trivial until you remember the first game sold 5 million copies by March 2021 despite mixed opinions on gameplay and story. People stayed partly because the walk looked worth it.

Gained: cheap multiplayer that feels like MMO without netcode bills.

Death Stranding calls itself a strand game built around social elements and asynchronous online functions. Other players never appear live; their ladders, ropes, roads and signs do. From an engine view this is brilliant thrift: no tick-rate war, no lag compensation, no 500-player shard. You get the emotion of cooperation — someone left a bridge — with the server cost of a message board. My take: this is the real reason a small newly independent studio could ship an online-feeling open world in 2019 and scale it in 2025. Decima did not need to become a netcode engine; it needed to become a mailbox engine.

Gained: shared R&D with Horizon.

Forbidden West used an updated version of Decima and went on to sell over 8.4 million units by April 2023. For Forbidden West, Decima even incorporated and built on the open-source Jolt physics engine rather than using a proprietary physics core. Kojima benefits when Guerrilla pays for terrain, animation and PlayStation optimization, just as Guerrilla benefits when Kojima stresses the engine with cargo physics and async structures. Friendship as middleware is fragile, but it is cheaper than building alone.

Cost: dependency and a late PC story.

Proprietary loans cut both ways. Kojima cannot hire from a large Decima labor pool the way Unreal studios can. Every PlayStation 5 Pro feature, like PSSR functionality noted for Death Stranding 2, has to be negotiated with Guerrilla's roadmap. The Windows version lag is visible: PS5 in June 2025, Windows in March 2026. Horizon had the same pattern, with Windows in March 2024 two years after PS4/PS5. Decima learned PC late.

Cost: not a shooter core.

Decima started in Killzone, but its modern identity is third-person traversal and machine combat, not crisp first-person gunfeel. Death Stranding 2's answer is telling: make weapons non-lethal by default to avoid voidouts and let players walk around fights. My reading: this is constraint turned into philosophy. If your engine makes walking prettier than shooting, reward walking.

My contrarian take: boring tech for boring work, on purpose

Most engine postmortems praise frame rate, destruction or lighting. I think Decima's contribution to Death Stranding 2 is more awkward: it makes weight readable.

Watch any delivery: Sam leans, stumbles, cargo shifts, Timefall chews containers. That legibility comes from years of Horizon work on traversal, cloth, terrain deformation and weather readability. Unreal could replicate it, but Guerrilla had already paid for it to make Aloy climb convincingly. Kojima essentially rented a climbing team to make a mailman.

Second, I think Kojima chose async because he distrusts live players, not because he loves them. Metal Gear Online taught him that live invasion breaks mood. The strand system keeps the generosity of multiplayer — roads appear — while deleting humiliation, griefing and voice chat. It is multiplayer for introverts, and Decima's world-persistence tools made it shippable.

Third, the Dejima story is not trivia. A Dutch trading post on a Japanese island is exactly what Kojima Productions Amsterdam was: a small foreign outpost allowed to trade technology inside a closed system. Sony published both sides, which helped, but the model does not scale. Indie devs should not expect Guerrilla to answer their email.

So here is the unfashionable lesson I would steal: pick the engine that makes your dullest loop legible, not the engine with the best demo. If your game is 80% carrying, pick the engine where carrying looks and feels precise. Everything else — combat, skill trees, day-night AI — can be thinner if the walk never lies.

What indie devs can actually copy

Prototype the trudge first. Before weapons, test slope, stamina, camera shake and cargo UI for 20 minutes with no enemies. If that is boring in graybox, no engine will save it.
Fake MMO with mailboxes. You do not need real-time netcode to get social warmth. Ghost structures, shared signs, likes and lost-cargo returns give 70% of the feeling for 10% of the server cost. This is directly visible in how lost cargo from other players can be recovered to complete deliveries.
Default to non-lethal. If your AI and animation budget is small, do what Death Stranding 2 does by defaulting to non-lethal tools. You avoid fail-state rage, you justify simpler enemy behavior, and you turn combat into puzzle routing.
Borrow, but price the dependency. A proprietary loan is not free. Ask who fixes bugs when the owner ships their own sequel, how PC ports are staffed, and whether physics middleware like Jolt or Havok will change under you. Decima's history shows both Havok and Jolt phases; plan for middleware churn.

Train the same muscles in a browser

You cannot practice Death Stranding 2's balance model directly in a tab, but you can train the underlying judgments: line choice on bad ground, infrastructure thinking, and open-world vehicle discipline.

If you want to feel why route beats speed, try MX Offroad Master. It is a 3D downhill mountain biking game about steep, bumpy trails where braking early keeps the bike stable. The Mountain Ride mode has 14 downhill levels across winter and spring maps, while Free Ride is an open practice arena for jumps and landings. That split mirrors Death Stranding 2 well: push timed deliveries when you must, but learn braking, steering and jump control in a no-timer space first. The rule that sudden full-speed charges into rocks and drops usually end in a crash is basically the porter rule.

If you want to practice the building half — roads, bridges, monorails — load Mine Clone. It is a blocky creative playground where every block can be broken, collected and reused. Start in Creative Mode with infinite resources to sketch a network, then switch to Survival Mode where gathering and night threats force tradeoffs. My suggestion: start small with one shelter, gather wisely, then plan a road before you need it. That is the Chiral Network in miniature: infrastructure first, deliveries second.

If you want open-world driving without the apocalypse pressure, use Sandbox City - Cars, Zombies, Ragdolls. It is an open-world zombie sandbox where you roam, drive cars, fight on foot and complete missions for cash. Walk up and press E to enter or exit cars, use cars to cross the city fast and ram groups on wide roads, then get out to loot cash and ammo on foot. Turn on Disable Zombies in Settings for combat-free driving and looting when you want a quiet delivery run. That peaceful-cruise mode is closer to Death Stranding's empty America than any shooter is.

None of these are Decima, and none prove you can ship a strand game. They are useful because they isolate one judgment at a time: ground reading, building order, and vehicle-versus-foot discipline.

FAQ: the engine questions people actually type

Is Death Stranding 2 on Unreal Engine 5?

No. Its engine is listed as Decima, with an improved version for the sequel.

Is it still Fox Engine?

No. Fox Engine stayed with Konami and was later dropped for Unreal on new hardware. Kojima Productions started fresh after 2015 and partnered with Guerrilla.

Why not switch to Unreal for the sequel?

My analysis: switching would have thrown away six years of traversal tuning, async-structure tooling and PlayStation-specific optimization shared with Horizon. Unreal hires easier and ports faster, but it would have forced Kojima to rebuild the walk — the one thing players actually buy.

Will Decima stay PlayStation-only?

No. Decima now spans PlayStation, Windows, macOS, iOS, iPadOS and Xbox Series X/S across its game list, even if individual releases stagger. Death Stranding 2 follows that stagger with PS5 first, Windows later.

Death Stranding 2 bet that walking is content if the ground, weather and other players' leftovers are readable enough. Decima, a borrowed Dutch engine named after a trading island, was the cheapest way to buy six years of that readability in 2016 — and the most stubborn way to keep it in 2025.

Why Alan Wake 2 Bet on Northlight: The Stubborn Finnish Engine Behind Remedy's Horror Comeback

dev game — Thu, 22 Oct 2026 08:07:38 +0000

Alan Wake 2 runs on Remedy's own Northlight Engine, not Unreal or Unity. That choice explains both why its flashlight combat feels physical and why it skipped last-generation consoles.

Remedy developed the 2023 survival-horror sequel with Epic Games Publishing for PlayStation 5, Windows, and Xbox Series X/S, releasing on 27 October 2023 with Northlight listed as its engine. The game sold over 2 million units by December 2024, making it Remedy's fastest-selling game, on a reported development and marketing budget around 70 million euros. Alan Wake 2

The same engine powered Control in 2019 and Quantum Break in 2016. Control was built in three years on about a 30 million euro budget, shipped first for PlayStation 4, Windows, and Xbox One on 27 August 2019, and later passed 6 million units by March 2026. It was explicitly described as using Remedy's in-house Northlight Engine and as among the first games to use hardware-accelerated real-time ray tracing. Control Quantum Break was the debut of that new engine, built with Digital Molecular Matter destruction, released 5 April 2016 for Windows and Xbox One. Quantum Break

Remedy itself says Northlight powers Quantum Break, CrossfireX story mode, Control, Alan Wake 2, and the next Control games, with a plugin-based tools framework, USD adoption, a modern game-object model, and a GPU-based geometry pipeline. The studio frames in-house work as control over the entire tech stack through close collaboration between creative and technology teams. Northlight

Snapshot: game, tech, and scale

Games: Alan Wake 2 (27 Oct 2023, PS5 / Windows / Xbox Series X/S), Control (27 Aug 2019, PS4 / Windows / Xbox One plus later PS5 / Xbox Series X/S ports), Quantum Break (5 Apr 2016, Windows / Xbox One)
Engine: Northlight, proprietary Remedy technology
Team context: Remedy was founded in August 1995 by members of demoscene group Future Crew, later created the Final Reality benchmark before spinning it off as Futuremark, and now employs 387 people as of 2026
Commercial result: Alan Wake 2 over 2M by Dec 2024, fastest seller; Control over 6M by Mar 2026; Alan Wake 2 budget around 70M euros

The demoscene origin matters more than it looks. This is a studio that started by squeezing effects out of home computers in a basement in Espoo, then built benchmark software to prove it. Keeping a custom engine is not just a technical decision. It is cultural continuity.

Why not Unreal? Control over light, destruction, and tools

Most horror teams license Unreal for proven lighting, Nanite geometry, and hiring convenience. Remedy refused that trade for three practical reasons.

1. Light is gameplay, not decoration.
In Alan Wake 2, Saga Anderson and Alan Wake fight with firearms plus a focusable flashlight. Focusing the light makes enemies vulnerable, but it drains the battery, forcing players to ration batteries and ammo and to dodge at close range. The Mind Place for Saga is described as a 3D menu where you pin clues on a board to solve the mystery, while Alan's Writer's Room lets you rewrite plot details to reshape the space around him. Alan Wake 2

A licensed engine would treat that flashlight as another spotlight and that menu as UI overlay. Northlight treats both as world systems. Shadows must be stable enough to aim by, batteries must read clearly at the edge of vision, and manuscript pages must foreshadow events without breaking streaming. When your core verb is burn darkness off an enemy, you want to own the light solver.

2. Destruction and shifting architecture need a bespoke object model.
Control is set in the Oldest House, a Brutalist headquarters with an interior far larger than its exterior that constantly shifts. The building was designed to show off destructible environments, and Jesse Faden's powers — Launching rubble, Shielding with debris, Levitation, Seizing enemies — all assume concrete can become ammunition. Control

Remedy's current pitch for Northlight is exactly this: an efficient game-object model plus GPU-based geometry for expanding virtual worlds, with customizable USD pipelines. Northlight That is not marketing fluff. It is how a mid-sized team can let players tear an office apart without the engine choking on draw calls.

3. Ray tracing early, path tracing later, on their own schedule.
Control shipped as one of the first hardware ray-traced games. Alan Wake 2 pushed further into mesh shaders and heavy path-traced lighting, which is why it launched current-generation only. Staying on Northlight let Remedy adopt DirectX Raytracing when Control needed reflections and then double down on dense forest lighting for Bright Falls and Watery when Alan Wake 2 needed fog, flashlights, and lake murk — without waiting for a vendor roadmap.

The official Alan Wake site still sells the fantasy the engine has to sustain: 13 years trapped in the Dark Place where nightmares manifest as reality, plus three Pacific Northwest locations — Bright Falls, Watery, and fenced-off Cauldron Lake. Alan Wake 2 official story Fog that thick is expensive. Remedy chose to pay the GPU cost rather than fake it.

What it cost: money, machines, and reach

The bill is public and sobering.

Alan Wake 2 reportedly cost about 70 million euros to develop and market, making it one of the most expensive Finnish cultural products. Control cost about 30 million euros over three years. Remedy absorbed the difference by staying mid-sized, reusing technology across the Remedy Connected Universe, and eventually self-publishing Control in 2025 after the 505 Games era.

Players paid a different price: no PlayStation 4 or Xbox One version of Alan Wake 2, steep PC requirements at launch, and a reputation for being a hardware punisher. Quantum Break's Windows 10 version was already criticized for technical issues at launch. Quantum Break Northlight buys visual coherence at the cost of forgiveness. If your PC is weak, the darkness that is supposed to scare you just stutters.

There is also creative risk. Two expansions — Night Springs on 8 June 2024 and The Lake House on 22 Oct 2024 — extended Alan Wake 2, while Control needed two expansions including the AWE crossover to fully tie the shared universe together. Alan Wake 2 A custom engine makes that kind of cross-game storytelling cheaper in the long run, but only if you survive the first launch.

My reading: three contrarian takeaways

These are interpretations, not Remedy statements. I label them as such because the task calls for distinct thinking, not another recap.

1. Demoscene pride beats license math.
Most analyses frame custom engines as NIH syndrome. I think Remedy keeps Northlight for the opposite reason: hiring and identity. A studio founded by Future Crew members that built Final Reality does not want to be another Unreal shop. Owning the renderer lets a 387-person team in Espoo and Stockholm hire people who want to write GPU code, not blueprint shoppers. The engine is a recruiting filter disguised as technology.

2. Northlight is a flashlight engine, not a landscape engine.
RAGE streams cities, Decima hikes mountains, Creation Engine hoards objects. Northlight renders attention. The flashlight cone, the Mind Place pinboard, the Writer's Room outline — all three turn looking into a mechanic. That is why Alan Wake 2 could switch from action-adventure to full survival horror without changing engines. The tech was already about scarcity of vision. Survival horror just turned the brightness down further.

3. The Connected Universe is engine amortization with lore.
Jesse Faden cameoing via the Federal Bureau of Control, Agent Kiran Estevez arresting Alan, the AWE expansion linking Control to Alan Wake — fans read this as Marvel-style planning. I read it as asset strategy. Once you have paid for destructible Brutalism and flashlight volumetrics, every new story that reuses the Oldest House, Cauldron Lake fog, or light-sensitive enemies makes the 70 million euros cheaper per scare. The spiral, not loop ending of Alan Wake 2 is a good metaphor for Northlight development itself: same place, slightly higher each time.

What indie devs can actually steal

You do not need a GPU cluster to copy the useful parts:

Make one light matter. Give the player a single rechargeable tool that gates damage, like the focusable flashlight that drains its battery. Scarcity of light is cheaper than dozens of enemy types.
Ship a diegetic notebook. Saga's Mind Place works because it never leaves the fiction — a 3D menu where pinning clues is progress. Even a Twine game can do this with a corkboard screen.
Constrain architecture to enable destruction. Control chose Brutalism because boxes break cleanly. If you want physics on a budget, pick concrete offices over forests. Fewer shaders, more rubble.
Plan for a second game on the same tech. Remedy spread Northlight cost across Quantum Break to Control to Alan Wake 2 to FBC: Firebreak and the upcoming Control Resonant. Scope your first game so its lighting or destruction system can carry the next one.

Train the same muscles in a browser

You cannot practice Northlight without a dev kit, but you can practice its verbs — light discipline, clue linking, and quiet searching — for free:

For light-and-sound stealth, try Horror Tale 1: Kidnapper. You escape a barn, find a missing bike wheel, crouch with C to stay quiet, throw objects to distract the patrol, and pick Ghost mode to learn routes before Hard mode punishes noise. Alan Wake rations batteries; you ration footsteps. The muscle is the same.
For Mind Place thinking, try Rooms Home Escape. Every drawer, painting, and locked cabinet hides part of a multi-room chain, and the guide explicitly tells you to jot down codes and revisit old rooms when new clues reframe them. That is Saga's pinboard without the FBI badge.
For house-scale dread under pressure, try 911: Prey. You play a kidnapped teenager escaping a maniac's house full of cameras and traps, where covering tracks matters as much as finding keys. Crouch, time moves, and do not leave drawers open — a low-fi version of Cauldron Lake tension.

None of these replicate path tracing. All three teach the loop Northlight was built to serve: observe, link, then move quietly through darkness.

FAQ

What engine does Alan Wake 2 use?
Northlight, Remedy's in-house engine. It is listed alongside Remedy as developer and Epic Games Publishing as publisher for the 27 Oct 2023 release. Alan Wake 2

Why didn't Remedy use Unreal Engine 5?
Remedy prioritized owning its lighting, destruction, and USD tools pipeline over hiring speed. Northlight already carried Quantum Break and Control forward, so Alan Wake 2 inherited ray-tracing and GPU geometry work instead of restarting on a vendor schedule. Northlight

Is Northlight only for horror?
No. It shipped a time-stutter shooter (Quantum Break), an open Metroidvania-style action game (Control), and survival horror (Alan Wake 2). Horror just uses its strengths — focused light and shifting interiors — most visibly.

Will my low-end PC run Alan Wake 2?
Poorly at launch settings. The current-gen-only scope and heavy lighting are the direct cost of the Northlight bet. Lower resolution, update drivers, and expect fog and flashlight volumetrics to remain the bottleneck.

Scope: PlayStation 5, Windows, and Xbox Series X/S versions of Alan Wake 2 as of Oct 2023, plus Control and Quantum Break history for context. Sales and budget figures are as reported by Dec 2024 to Mar 2026 sources and will shift with re-releases and Control Resonant.

Why Forza Horizon 5 Bet on ForzaTech: The Boring House Engine Behind a 10-Million-in-a-Week Mexican Festival

dev game — Thu, 22 Oct 2026 04:37:01 +0000

Forza Horizon 5 did not win because it had the prettiest cars. It won because Playground Games refused to rent an engine and kept living inside Microsoft's boring house engine, ForzaTech.

That house engine — built by Turn 10 Studios since 2001, trademarked by Microsoft in 2015, and now also powering the Fable reboot — let a British arcade team ship a 50%-bigger Mexico with 799 licensed cars, cloud-trained opponents, and 320 hand-recorded engine sounds, then survive the sim side collapsing underneath it. ForzaTech is the proprietary engine created by Turn 10 and used for current Forza games and the Fable reboot.

This is the first racing-engine article in this series: not Unity arcade physics, not Unreal Nanite density, but a 20-year-old car-culture pipeline that treats licensing, audio, and AI traffic as the engine.

The short answer: why ForzaTech instead of Unreal

Playground Games, founded in Leamington Spa by ex-Codemasters Dirt and TOCA veterans and inspired by Test Drive Unlimited's open world, pitched Turn 10 an open-world Forza at E3 2010 and built Horizon in close collaboration with Turn 10 before Microsoft acquired Playground in 2018. Playground was co-founded by Gavin Raeburn, Trevor Williams and Ralph Fulton, offered Turn 10 an open-world concept, and was acquired in 2018.

Unreal could have rendered Mexico. It could not have cheaply solved Forza's real bottlenecks:

799 licensable, tunable, paintable cars as of May 2026 with engine swaps, drivetrain swaps and body kits
Opponents that feel like your friends without paying for 100-player servers
Car sound that does not loop like a vacuum cleaner

So Horizon 5 stayed on ForzaTech. The game features 799 obtainable licensed vehicles as of May 2026, plus engine swaps and body kits. That roster size is the tell. You do not maintain that on a rented engine unless you also rent the car pipeline.

What ForzaTech actually is

Turn 10 was created in 2001 under Microsoft Game Studios to build an Xbox rival to Gran Turismo for PlayStation. Early staff had published Project Gotham Racing and Golf 4.0 but had not built games before, so they built process instead of genius. Turn 10 was established in 2001 to develop racing games that became Forza as the Xbox rival to Gran Turismo.

Three house tools matter more than graphics:

1. Fuel, the car factory. Turn 10 builds cars and tracks with a proprietary 3D modelling tool called Fuel that lets multiple artists work on the same model at once. For Xbox One cars it still took six months for four people to finish one car, so Turn 10 came to rely on manufacturers sharing CAD files, scanning real cars, or sending photographers to shoot thousands of photos of new launches. Turn 10 uses proprietary Fuel and relied on CAD files, scans and thousands of photos after six-month builds. This is why my first thesis below calls ForzaTech a licensing moat, not a renderer.

2. Drivatars, the Cambridge cloud. Every Forza title uses Drivatars, a portmanteau of driver and avatar, designed by Microsoft Research Cambridge. Early games used a Bayesian neural network that picked the highest-confidence turn angle and throttle from past races. Since Forza Motorsport 5 they use reinforcement learning on cloud racing data from all connected players, tracking position, speed and consistency to infer lines for tracks and cars the player never drove, timestamping uploads and down-weighting old ones. Turn 10 says only the cars are rubber-banded for distance, not the AI inputs, and it keeps a control layer between player and AI to block weird results. Drivatars were designed by Microsoft Research Cambridge, moved from Bayesian to reinforcement learning with cloud data since Forza Motorsport 5, with timestamping and a control layer.

3. ForzaTech as shared platform. The engine was trademarked in 2015 and now underpins both Motorsport and Horizon plus Fable. ForzaTech was trademarked in 2015 and has since been used for current Forza games and the Fable reboot. The reboot Fable lists ForzaTech as its engine, which proves Horizon work is not throwaway: open-world streaming built for Mexico is being reused for Albion.

Turn 10's own reboot shows both sides of that platform. Forza Motorsport (2023) lists ForzaTech as its engine, launched October 10, 2023 with 500+ vehicles, 800 upgrades and 20 tracks, with real-time ray tracing, tire wear, dynamic time, damage with paint chipping and dirt, animated track life, physically based lighting, volumetric fog and a procedural cloud system. It won Best Sports/Racing at The Game Awards 2023 and Racing Game of the Year at the 27th D.I.C.E. Awards, but players on Steam gave it mostly negative reviews over bugs, the Car Points RPG progression and thin content. It won those two racing awards while receiving mostly negative Steam player reviews for bugs and progression.

What Horizon 5 asked that engine to do

Horizon 5 is a 2021 racing game developed by Playground Games and published by Xbox Game Studios, fifth Horizon and twelfth Forza main entry, set in fictionalised Mexico, released November 9, 2021 for Windows, Xbox One and Xbox Series X/S, and for PlayStation 5 on April 29, 2025 by Panic Button — the first mainline Forza on a non-Microsoft platform. It released those dates on those platforms with Panic Button handling PS5 as the first non-Xbox Forza.

Commercially it launched to over 10 million players in its first week, the biggest-ever launch for an Xbox Game Studios game, won IGN's Game of the Year, three jury-voted awards at The Game Awards 2021 tying It Takes Two, and Racing Game of the Year at the 25th D.I.C.E. Awards. It hit 10 million in week one, won IGN GOTY, three Game Awards 2021 and D.I.C.E. racing.

Technically the asks were brutal:

Largest Horizon map ever, 50% larger than Horizon 4, with the series' highest point, described as one of the most diverse maps built, containing caldera volcano, jungles, beaches, Mayan temples and Guanajuato streets. The map is 50% larger than Horizon 4 with caldera, jungles, beaches, temples and Guanajuato.
Photogrammetry everywhere, built with Mexican artists for murals, Mexican musicians for soundtracks, and a field team capturing real light and sky data, down to cholla cactus needles visible only on Xbox Series consoles. The team partnered with Mexican artists and musicians, captured real light and sky, and used photogrammetry heavily including cactus needles on Series consoles.
Local weather instead of global weather: storm visible on one side of the map, four seasons affecting eleven biomes, dust storms in dry season, tropical storms in fall, jungle leaves reacting to wind. Local weather, four seasons over eleven biomes, dust and tropical storms, and reacting jungle foliage.
First in franchise to support ray tracing on cars, plus new Horizon Arcade pinata pops, EventLab custom games and races, AI assistant Forza Link that tracks player status to link co-op and GPS, over 1,800 Accolades with some cars locked behind them, returning Eliminator battle royale starting in a 1963 Beetle, and buyable houses from cabanas to a hotel enabling Wheelspins and fast travel. First ray tracing on cars, plus Arcade, EventLab, Forza Link AI, 1,800 Accolades, Eliminator Beetle start, and houses.
Sound rebuilt from loops to granular synthesis with over 320 unique engine sweeps from idle to redline to standstill, using flat-response padded mics to avoid exhaust clipping plus stereo and ambisonic cabin mics, building a bespoke library rather than reusing stock, output through Dolby Atmos. Transition to granular synthesis with 320 sweeps, padded flat-response mics and ambisonic cabin capture for a bespoke library.

Xbox's own store page frames the same pillars as diverse, adventurous, evolving, social and your open world: living deserts to snow-capped volcano, hundreds of Horizon Story challenges, weekly changing seasons with dust and tropical storms, Horizon Arcade with no menus or lobbies, and EventLab plus Gift Drops. Xbox lists those five open-world pillars for Horizon 5. Steam adds the mundane proof of scale: 110 GB storage, Windows 10 18362+, i5-4460 or Ryzen 3 1200 minimum with GTX 970, RX 470 or Arc A380, 16 GB and GTX 1070 recommended, plus a dozen car packs from Italian Exotics to JDM Jewels and Nissan Retro Rides. Steam lists 110 GB, those CPU/GPU minimums and the long car-pack tail.

Expansions kept the engine fed: Hot Wheels July 19, 2022, Rally Adventure with new map and trophy trucks March 29, 2023, and Fast X car pack December 5, 2023. Hot Wheels, Rally Adventure and Fast X dates and contents.

What it cost — and why the arcade side inherited the engine

This is where ForzaTech stops being a tech story and becomes an org story.

Turn 10 employed about 100 full-time plus 100-200 contractors in 2014, skipped Motorsport in 2019 to innovate, then shipped the 2023 reboot into mixed reception. In July 2025 Microsoft cut about 70 Turn 10 roles as part of 9,000 layoffs, reports said Motorsport was cancelled and remaining staff would support Horizon, support stopped at year end with servers staying online, and in September 2026 Turn 10 merged into Playground Games. Turn 10 had 100 staff in 2014, cut 70 roles in July 2025, stopped Motorsport content with servers online, and merged into Playground in September 2026. Playground itself had grown to 455 people in 2024, opened a second studio in 2017 for the non-racing open-world project that became Fable and a third in 2024 to help with Fable. Playground grew to 455 in 2024 with second and third studios for Fable.

My read — and this is interpretation, not Microsoft's statement — is that the sim saved the tech, then the arcade saved the tech's owners. ForzaTech outlived its house. Motorsport is now described as a parked asset, not discontinued, leaving long-term re-engagement open, while Horizon 6 carries the engine forward. Microsoft leadership calls Motorsport a parked asset rather than discontinued.

Horizon 6 is listed by Xbox for May 19, 2026 on Series X/S and Windows 10/11, set in Japan with Tokyo as Horizon's largest ever city and most dense map, over 550 real-world cars including JDM classics, Touge Battles, Aftermarket Cars, buyable homes with customizable garages, and CoLab, the multiplayer-capable EventLab upgrade. Xbox lists May 19, 2026, Japan/Tokyo largest city, 550+ cars with JDM and Touge, and CoLab. Fable, also on ForzaTech, is currently listed for February 23, 2027 with style-weaving melee/ranged/magic and a Living Population of 1,000+ NPCs with personalities and routines that continue while away. Fable on ForzaTech lists style-weaving and 1,000+ Living Population NPCs.

If you only remember one sentence: the festival game became the engine owner.

My analysis: four contrarian takes (labeled as opinion)

Other roundups will tell you ForzaTech wins on graphics. I think that misses why a house engine survived Unreal.

1. Boring is the moat. Fuel plus CAD plus thousands of reference photos is not sexy, but it is uncopyable quickly. Anyone can buy Unreal vehicle physics. Almost no indie can license 799 cars and keep each one's tune, livery and upgrade path coherent across two sub-series. My take: Playground bet that content pipeline beats renderer novelty. The 320 granular sweeps point the same way — bespoke audio library instead of stock loops is expensive once, cheap forever, and players feel it in every tunnel. Facts above on Fuel/CAD and 320 sweeps support the cost; the moat claim is mine.

2. Mexico was Fable pre-production. Eleven biomes, living deserts to snow volcano, local weather, photogrammetry rocks you can inspect on the caldera — that is an RPG biome test disguised as a racing map. Playground said its open-world design, technology and streaming could transfer to another game when staffing Fable, hiring GTA V, Metal Gear Solid and Hellblade devs for the RPG. My take: Horizon 5 was allowed to be 50% bigger partly to prove ForzaTech streaming for a fantasy open world. The hiring and engine reuse are fact; the disguised-test motive is my inference.

3. Drivatars are cheap social, not smart AI. An empty open world kills a festival fantasy, but 100-player synced racing is expensive and laggy. Cloud Drivatars that mimic friends' braking and lines give you recognizable rivals with timestamps and confidence decay, without servers simulating 799 cars. My take: this is multiplayer illusion on a budget, and it works because racing needs ghosts, not chat. The Cambridge Bayesian-to-reinforcement history and timestamping are fact; cheap-social framing is mine.

4. The arcade rescued the sim — backwards from usual. Normally sim tech trickles down to arcade. Here Motorsport's 2023 reboot stumbled on Car Points progression and content gaps while Horizon kept adding Accolades, EventLab and expansions. Then half of Turn 10 left and the remainder was tasked to help Horizon 6. My take: festivals retain players better than lap-time ladders, so the engine's future is owned by the fun side. Layoffs, parked-asset language and merger are fact; the retention moral is mine and deliberately debatable.

If you disagree, good — that is the point. House engines force you to pick which boring problem you will own for a decade.

What indie devs can steal without 455 people

Own your bottleneck tool, even tiny. You cannot copy Fuel, but you can copy the idea: one shared editor where art and design touch the same car/track/level live. A Google Sheet that exports to your game counts if everyone actually uses it.
Copy players, not AI. If your game has any repeatable movement — delivery routes, patrols, racing lines — record friends' runs as opponents before writing behavior trees. Drivatar logic started Bayesian, not deep magic.
Record, do not loop. Even ten hand-recorded variations with pitch and load changes beat one perfect loop. The 320-sweep number matters less than the rule: capture idle-to-redline-to-stop as one performance.
Test biomes in one map. If you plan three environments, ship one map with three micro-climates first to prove streaming, culling and audio transitions. Mexico's eleven biomes in one festival is the big-budget version.
Plan engine succession. Ask who owns your tech if your sim side dies. Document build, licensing contacts and car/track equivalents so the arcade side can inherit it. Turn 10's merger is a warning in reverse.

Train like a Horizon driver in a browser (no download)

You cannot practice 799-car tuning in a tab, but you can practice the three skills Horizon rewards: close overtaking, stunt commitment, and upgrade order. These free browser trainers map cleanly:

For traffic reading and near-miss discipline, play Racing Limits. Its Career-to-Infinite flow, close-call bonuses, one-way versus two-way risk, and handling-first upgrade order — handling and brakes before engine and nitro — is the closest browser analogue to surviving Guanajuato traffic before pushing speed. Use Free Mode to learn gaps, then chase combos.
For stunt commitment and map learning, try Madalin Stunt Cars 2. With 34 cars unlocked from the start, three open maps, nitro on Shift, handbrake taps on Space and instant reset on R, it teaches the Horizon festival habit: learn one ramp well on Map 1, then use Map 2 for races and Map 3 for exploring, carrying speed into loops instead of hoping mid-air.
For boost timing and upgrade priorities, run Night City Racing. Its 16-track Racing mode against AI, 5-task Challenge mode and coin-farming Free Drive mirror Horizon's race-earn-upgrade loop, and its rule — engine, agility and brake first, then turbo and booster, rims last — is exactly how to approach Horizon's early tunes without wasting credits.
For open-world collecting as progression, cruise Crazy Cars. Its star-and-wrench economy for 17+ cars, 8 tracks plus drag strips and Mount Egg climb, and R-reset flow teaches the Accolade mindset: explore first, collect deliberately, then take each new unlock back to an old jump to see what is now reachable.

If you only try one, make it Racing Limits for clean passes and Madalin Stunt Cars 2 for joyful crashing — Horizon needs both.

Bottom line

Horizon 5 bet that a proprietary car pipeline plus player-copying AI plus hand-recorded sound would beat a prettier rented renderer, and 10 million players in a week plus a PS5 port four years later suggest it was right for this genre. The cost was real: the sim reboot that shared the engine stumbled, half its studio left, and the arcade studio inherited the tech for Japan and Albion.

For indies the lesson is narrower and more useful: do not ask which engine is best. Ask which boring bottleneck you can own for ten years — car licensing, opponent ghosts, or sound — and let the festival pay for the factory.

Space Waves Guide: Controls, 33 Levels, Modes and Why One-Button HTML5 Design Wins

dev game — Thu, 22 Oct 2026 04:06:00 +0000

Space Waves looks too simple to be interesting: hold to go up, release to go down. Then you crash into the ceiling in four seconds and realize the simplicity is a trap.

Space Waves is a neon arcade dodger where you steer a glowing arrow through tight space tunnels. The arrow scrolls forward by itself — you only control height. The standard Classic set has 33 handcrafted levels you can pick in any order, plus Endless distance runs and Race ghost races. It was released in April 2024 as a browser game for desktop, mobile and tablet, most recently listed as a Unity 6 web build on CrazyGames, and it runs as an HTML5 no-download page that works on school Chromebooks and phones where browser games are allowed.

That combination — one button, instant restart, and zero install — is exactly why it keeps showing up in unblocked searches alongside Geometry Dash Wave practice. This guide covers the verified controls and modes, then adds a technical take most guides skip: why one-button design is a perfect fit for HTML5/WebGL delivery.

The 30-second answer

PC: Hold Space, W, Up Arrow, or left mouse button to rise. Release to fall. P pauses. Controls are identical on CrazyGames and EasyHub.
Mobile / tablet / Chromebook: Touch and hold to rise, lift to fall. Trackpad click-hold or Spacebar works the same way. Landscape + fullscreen gives the widest view.
Goal: Reach the finish without hitting a wall, rotating cog, or thorn. Crashes restart instantly, and many versions show distance percentage so you can measure progress.
Levels: Start with green calm-face levels (wide tunnels), then move to yellow/orange, then red/nightmare faces for tight angles and faster scroll.
Modes: Classic for finishing levels, Endless for high score, Race for pressure against visible ghosts you cannot collide with.

If you only remember one skill, remember this: feather, don't hold. Short rapid taps beat long holds everywhere past the first green levels.

Controls are a PWM signal, not a joystick

Every guide says hold to go up, release to go down, which is true but undersells what you are actually doing. EasyHub's Space Waves guide describes the same rule plus the beginner mistake: holding too long slams the ceiling, releasing too long hits the floor, and straight tunnels require constant zigzag because the arrow flies a fixed diagonal and cannot hover.

Here is my take, and it changes how you practice:

Think of your finger as a PWM (pulse-width modulation) controller. You are not steering — you are setting a duty cycle. A 20% hold / 80% release drifts down. A 50/50 rapid tap holds mid-height. A 90% hold rockets up.

Why does this mental model help? Because it explains why mouse players often feel less consistent than Spacebar players. A mouse click has a mushier press-release edge than a key. If your clicks feel inconsistent, switch to Spacebar, play fullscreen, and close extra tabs to cut input lag — advice that matches the general HTML5 guidance to use fullscreen and a stable 60 Hz+ display for timing games.

On mobile, the same idea applies: small thumb taps, not long presses. Play in landscape with auto-rotate on. Long touches cause big climbs that you cannot recover from in narrow sections.

You can try the exact control feel free in your browser on Space Waves on EasyHub — no download, pick any of the 33 levels immediately to test feathering without grinding unlocks.

33 levels and what the faces actually mean

The Classic set is described as 33 levels across three worlds — simple early worlds to lava-floor later themes — with open select from the start. EasyHub uses the same 33-level count and explains faces and colors as a risk meter: green/calm for wide tunnels, yellow/orange for tighter gaps with spikes and cogs, red/angry/nightmare for very tight angles and faster scroll, similar to Demon-tier wave difficulty.

My contrarian read: open select is the real design win, not a casual concession.

Most arcade runners lock you into progression to manufacture retention. Space Waves lets you dodge a wall by switching levels. Stuck on a nightmare hinge? Drop to a green level for five minutes to rebuild rhythm, then return. That loop — hard attempt, easy reset, hard retry — keeps muscle memory warm without tilt. I think this is why the game works as Geometry Dash Wave training: it is not a clone, it is a wave-only practice sandbox with measurable levels instead of an endless grind.

Do not trust mirrors that list wildly different totals as fact. Some list extra worlds or mode tracks. Treat 33 as the standard Classic set and Endless/Race tracks as separate.

How to beat hard and nightmare levels

Hard levels are not a reaction test. They are a reading test. This method synthesizes CrazyGames tips on zigzag timing, staying low around circular spikes, and starting moves early with EasyHub's feathering, look-ahead, and recenter routine:

Feather in micro-taps. Long holds kill you in narrow sections. Shrink inputs as speed rises.
Look 2-3 obstacles ahead. Keep the arrow in peripheral vision. Start climbing or dropping early because response lags slightly.
Default to the center line. My term for this is center-line economy: after every dodge, return to middle. It buys you room to go either way. Dodge, recenter, dodge, recenter.
Find the hinge. Every level has one section causing most deaths. Spend runs learning just the opening rhythm, then isolate that hinge instead of doing full runs.
Memorize, don't react. Layouts repeat. After a crash, replay the spot mentally: where was the gap, was the saw on a cycle, was the safe line high or low?
Stay low around circular spikes. Trying to float over them usually ends on the ceiling. Thread under when the layout allows it.

If you start click-spamming, stop. Panic causes ceiling-floor bouncing. Take a short break, then return with one goal like surviving the first speed-up cleanly. Instant restarts make 20 focused retries more valuable than one tired hour.

Classic, Endless, Race — and why Race ghosts are clever tech

The three modes solve three different practice needs:

Classic: Pick a world and level, reach the finish. Best for short sessions and pattern learning.
Endless: No finish, chase distance while speed and density rise. Best for warm-up and score chasing.
Race: Finish against visible ghosts with a top tracker. You cannot bump them, but you see who crashes or pulls ahead. Easy and hard options suit different goals.

CrazyGames lists the same three modes and notes that in Race you see all other racers' movements and crashes on a tracker, but looking away for a second is enough to crash yourself. EasyHub adds that crashes restart instantly and some versions offer practice resumes from near your crash point.

Here is the technical angle other guides miss: Race ghosts are fake multiplayer that avoids real multiplayer cost.

True online racing needs WebSockets/WebRTC netcode, lag compensation, and servers — heavy for a lightweight browser game. Ghosts give you social pressure with zero netcode: just replay data or simulated paths. You get the tension of someone is ahead without the Unity WebGL build paying for networking, room matching, or anti-cheat. For an HTML5 instant-play title, that is the right tradeoff. Ignore the leaderboard while moving; check it only after you finish a section.

Practice mode with resume-from-crash is the same philosophy: reduce dead time. The faster you retry the hinge, the faster you learn.

Why one button wins on HTML5 / Unity WebGL

Space Waves is listed as a Unity 6 browser build, and Unity web exports boil down to a predictable set: a loader.js to bootstrap, framework.js runtime, .wasm WebAssembly binary, and .data assets. Unity's manual describes exactly that Build folder structure for web distribution.

The open web stack behind that is HTML5, WebGL for graphics, Touch Events and Fullscreen API for input/display, and local storage like IndexedDB for saves — the core web-is-the-platform set described by MDN.

My analysis: one-button games are structurally ideal for this stack, for three reasons guides rarely state:

Minimal input surface hides latency. Left/right plus jump plus build plus shoot needs precise key mapping and low jitter. Hold/release survives a 30-50 ms browser hiccup because you correct with the next tap. That is why Chromebook trackpads and phone touch still feel fair.
Tiny state, fast load. No inventory, no large maps, no voice chat. A neon tunnel plus arrow plus obstacle seeds compresses well into .wasm plus .data, loads like a normal page, and fits school filters that block big downloads.
Fullscreen is a performance feature, not cosmetics. Fewer composited tabs plus fullscreen plus 60 Hz reduces frame pacing noise that destroys diagonal-wave timing. Close tabs before nightmare attempts — it matters more than a new skin.

Skins (fiery arrow, rounded arrow, color variants) do not change physics. Pick what you can see best against neon backgrounds. Contrast beats style.

Unblocked access still depends on your network. If a school filter blocks a domain, no game setting bypasses the administrator. Prefer allowed browser versions, avoid unofficial app downloads, and use a modern browser with JavaScript enabled.

If you like Space Waves, train these adjacent skills

You do not need a new guide for every dodger — you need transferable reading skills. These three are natural next steps, all playable free in the browser:

For look-far-ahead reading under strobing speed, try Tunnel Rush on EasyHub. It is first-person tunnel dodging with left/right only, 100 structured levels, and the same lesson: focus on the vanishing point, use micro-taps, and recenter after every wall. If Space Waves taught you vertical feathering, Tunnel Rush teaches horizontal discipline.
For speed scaling and risk economy, try Crazy Roll 3D on EasyHub. Like Slope, you steer a ball down endless neon slopes as speed rises, but you add diamond lines, Shield/Magnet/2x timing, and a ball shop. My rule transfers directly: skip 30 percent of risky diamonds to protect the run — same as skipping a tight Space Waves gap to hold center.
For one-button release timing, try Stickman Hook on EasyHub. Hold to attach, release to fly — the inverse of Space Waves hold-to-rise. Players who master feathered release windows in Space Waves usually learn the release just past the bottom, on the way up swing window faster, because both punish holding a good move too long.

Each of those uses the same HTML5 instant-play advantage: open, fullscreen, retry in seconds.

Quick FAQ

How many levels are in Space Waves?
33 in the standard Classic set, plus Endless and Race tracks on versions that include them.

Do I have to unlock levels?
No. Pick any level from the start and use difficulty faces as your guide.

Why do I keep hitting the ceiling?
Holding too long. Switch to feathered micro-taps and recenter after each dodge.

Is Space Waves official Geometry Dash?
No. It is inspired by Geometry Dash Wave — hold to climb diagonally, release to dive — but it is a fan-style tribute focused only on wave flying, not an official RobTop game.

Best 15-minute practice loop?
Warm up with one green Classic level, do five hinge-only retries on your target nightmare level, finish with one Endless run for rhythm. Stop when you start spamming inputs.

Why Zelda Bet on Havok and a Homegrown Chemistry Engine: The Weak-Switch Physics Lab Behind 58M Copies

dev game — Thu, 22 Oct 2026 00:34:08 +0000

TL;DR

The Legend of Zelda: Breath of the Wild (2017) and Tears of the Kingdom (2023) run on Nintendo EPD's proprietary engine plus a modified version of Havok physics, not Unreal or Unity. That boring-sounding choice powers a second, homegrown layer Nintendo calls the chemistry engine. Together they sold 35.76 million (BOTW, by June 2026, the bestselling Zelda ever) and 22.71 million (TOTK, by June 2026, including 10 million in three days) on hardware that peaks at only 197 GFLOPS undocked. BOTW Wikipedia TOTK Wikipedia Switch specs

Most explainers stop at "great physics, very emergent." My argument is sharper: Nintendo used weak hardware as a design filter, split simulation from meaning, and shipped a verb-first sequel. TOTK reuses the same Hyrule map but triples what you can do — Ultrahand, Fuse, Ascend, Recall, Autobuild — instead of building a bigger map. If you are an indie dev, that is the lesson to steal.

The bet in one table

Question	Zelda answer	What AAA usually does
Engine	Nintendo EPD proprietary + modified Havok	Unreal Engine 5
Second system	Homegrown chemistry engine (fire, wind, electricity, temperature, material tags)	Scripted triggers
Hardware target	Wii U / Switch (Tegra X1, 4 GB LPDDR4, 197 GFLOPS undocked, 393 GFLOPS docked)	PS5 / high-end PC first
Map strategy	Reuse Hyrule, add Sky + Depths + construction verbs	New, larger map
Prototype	Playable 2D Zelda-like to test physics puzzles first	Vertical slice with final art

Nintendo EPD led both games under director Hidemaro Fujibayashi and producer Eiji Aonuma after a five-year rethink that started right after Skyward Sword (2011). Monolith Soft, veterans of Xenoblade open worlds, assisted with landscape and topography. BOTW development

What Havok actually does (and does not do)

The final game uses a modified version of the Havok physics engine. That line matters because people over-credit Havok. BOTW development

Havok is middleware for real-time collision detection and rigid-body dynamics, now owned by Microsoft since 2015 and explicitly optimized for Nintendo Switch. Havok's own pitch is telling: it is made to be embedded in countless proprietary engines. Havok software Havok official

In Zelda, Havok handles the dumb, honest work:

boxes slide, boulders roll, ragdolls tumble
Ultrahand assemblies hold together or shear apart
paraglider lift, shield-snowboarding friction, stasis launches conserve momentum

Havok does not decide what fire means. That is the chemistry engine.

The chemistry engine is the real engine

BOTW introduced a consistent physics engine plus a chemistry engine that defines the physical properties of most objects and governs how they interact. BOTW gameplay

Concrete examples from the game:

metal attracts lightning in thunderstorms — drop a sword near enemies and let the sky do the work
wood burns, spreads fire, creates updrafts; grass + wind + fire = a flying machine
cold + heat + food = timed buffs; wooden shields catch arrows but also catch fire
rain makes cliffs slippery, which quietly nerfs climbing and forces a new plan

My framing, clearly labeled as opinion: Havok simulates, chemistry interprets. Havok says "this box moved." Chemistry says "this box is wooden, wet, electrified, on fire, and uphill from dry grass." That separation is why Zelda feels systemic while many Unreal open worlds feel scripted. Unreal can do all this, but Nintendo built the meaning table first and let Havok do the math.

At GDC 2017 the team called this Change and Constant — Breaking Conventions. They showed the 2D prototype they used to prove physics puzzles were fun before any 3D art existed, and Aonuma recalled walking into a test area to find everything blown away by wind. That bug became a pillar: keep the world legible, keep wind real. BOTW development

Why not Unreal or Unity? The 197-GFLOPS answer

Switch is a tablet: Nvidia Tegra X1, quad-core ARM Cortex-A57 at 1.02 GHz, 256 Maxwell CUDA cores, 4 GB RAM. Docked you get 393 GFLOPS; undocked you get about 197 GFLOPS. Switch hardware

On that budget, photoreal Nanite-style geometry is suicide. Nintendo's trade was:

Gain: a stable sandbox where every object follows the same 10 rules, so players can invent solutions (octo-balloon rafts, mine-cart catapults, log bridges).

Cost: blurry textures, pop-in, frame-rate criticism that followed both games, especially TOTK. Reviewers praised scale and mechanics while flagging frame rate. TOTK reception

Here is my contrarian take: weakness was a filter, not just a limit. Because they could not win on fidelity, they had to win on interaction density. Most AAA open worlds add square kilometers; Nintendo added verbs. TOTK's concepts even started as BOTW DLC ideas that exceeded scope — instead of a new map, they built Ultrahand, Fuse, Ascend, Recall, Autobuild plus Zonai devices (fans, rockets, sleds) and two new layers, sky islands and the Depths. TOTK development

That is a verb-first sequel. Same Hyrule, triple the grammar. I think this is the opposite of industry wisdom, and I think Nintendo is right. Players remember what they did (fused a rocket to a shield to fly), not how many acres they crossed.

Two other unpopular opinions:

Weapon durability is memory management disguised as philosophy. Breakable weapons force rotation. If you could hoard one perfect sword, you would stop experimenting with fire, magnetism, and gravity. Breaking is the garbage collector for creativity.
Loose Ultrahand snapping is intentional bad UX. Strict grid snapping would look cleaner but kill comedy. Wobbly, slightly-off vehicles that collapse hilariously teach statics better than a tutorial. Nintendo preserved failure because failure is searchable — clips of failed mechs sold more copies than trailers.

BOTW vs TOTK: what changed in the stack

BOTW gave you Sheikah Slate powers: remote bombs, Magnesis, Cryonis, Stasis. You could climb nearly anything, glide with the paraglider, cook for buffs, and trade Spirit Orbs for hearts/stamina. BOTW gameplay

TOTK keeps that open-world foundation but replaces Slate powers with five construction verbs: Ultrahand (grab + glue), Fuse (weapon + material), Ascend (swim up through ceilings), Recall (rewind motion), Autobuild (reprint a saved design, spending zonaite if parts are missing). Add Korok seeds, shrines, Lights of Blessing, and Zonai parts, and you get a survival-crafting game inside a Zelda game. TOTK gameplay

Tech consequence nobody talks about: Autobuild is basically version control for physics. It serializes your contraption, then re-materializes it. That only works because every part has clean chemistry tags + Havok constraints. Indie takeaway: if your objects do not have stable IDs and material tags, you cannot do "save my build."

TOTK also shipped enhanced Switch 2 editions in June 2025 for both games, proving the stack scales up when hardware finally arrives. BOTW release TOTK release

What indie devs can steal this weekend

You do not need a Havok license. Use Jolt, Rapier, or even Box2D + a tiny chemistry table.

Prototype in 2D like Nintendo did. One screen, one magnet, one fire tile, one metal box. If it is not fun flat, 3D art will not save it.
Write a 10-row chemistry table before you write quests. Example: wood=flammable, metal=conductive+magnetic, ice=meltable, wind=force. Every puzzle is then a row combination, not a script.
Budget verbs, not acres. One new verb (glue, rewind, ascend) beats 2x map size for clips and retention.
Make breaking generous. Durability, zonaite costs, stamina — all force tradeoffs. Give players Autobuild-style favorites so loss stings less but still teaches.
Keep physics legible at 30 fps. Cap rigid bodies, sleep idle objects, use simple colliders. BOTW's readability (bright outline, clear sound when Ultrahand connects) matters more than accuracy.

Train the same muscles in your browser

You cannot practice Ultrahand without a Switch, but you can practice the thinking: fast building, resource budgeting, and treating physics as a tool.

If you like TOTK's freeform construction, warm up with quick-build duels in BuildNow GG. Its Freebuild island is perfect for drilling 90s, box-peeks, and ramp rushes — the same snap-glue-edit loop as Zonai building, just faster and with shotguns.

For slower, Minecraft-style planning, try Mine Clone. Creative vs Survival forces the same choice TOTK forces with zonaite: infinite imagination versus budgeted gathering, plus day-night triage.

If you want cozy, detailed building with 300+ blocks and furniture, Craft World trains the decorator half of TOTK — layering textures, using vertical space, and rebuilding without penalty.

And for pure emergent messing around — ram zombies on wide roads, hop out to loot, toggle slow-mo for stunts — Sandbox City is the closest browser vibe to BOTW's "no single forced path" Hyrule. Drive to a hotspot, park, finish on foot, reposition by car: that is exactly how I clear BOTW camps with boulders and lightning.

None of these replace Hyrule, but together they build construction speed, material intuition, and a habit of asking "what if I glue this to that?" instead of "where is the intended solution?"

FAQ

Did Zelda use Unreal Engine?
No. Both games use Nintendo EPD proprietary tech plus modified Havok physics. There is no Unreal or Unity in the core loop. BOTW development

What is the chemistry engine?
Nintendo's internal ruleset that gives objects material properties (wood, metal, fire, ice, electricity) and lets them interact consistently, on top of Havok rigid-body simulation. BOTW gameplay

How big were they really?
BOTW: March 3, 2017 (Wii U + Switch launch), 35.76M by June 2026, Game of the Year at The Game Awards 2017 and D.I.C.E. TOTK: May 12, 2023, 10M in 3 days, 22.71M by June 2026. BOTW sales TOTK sales

Why does TOTK reuse the same map?
My read: reusing Hyrule let EPD spend budget on verbs (Ultrahand, Fuse, Ascend, Recall, Autobuild) and vertical layers (sky + Depths) instead of terrain. Familiar geography makes new powers feel stronger because you can compare old vs new routes.

Bottom line

Unreal sells you fidelity. Nintendo EPD + Havok + chemistry sells you consequences. On a 197-GFLOPS handheld, that was not nostalgia — it was arithmetic. Build fewer acres, build more meanings, prototype flat, let wind blow your level away once, and keep the part where players surprise you.

Why Battlefield 6 Bet on Frostbite: The Difficult House Engine Behind 7 Million Copies in 3 Days

dev game — Thu, 22 Oct 2026 00:04:07 +0000

Battlefield 6 runs on Frostbite, not Unreal Engine 5. That surprises outsiders, because Frostbite has a reputation as EA's difficult in-house engine — the same tech that slowed down Mass Effect: Andromeda, Anthem, and Dragon Age: Inquisition.

For Battlefield, the bet makes sense. Developer Battlefield Studios — led by DICE with Criterion Games, Motive Studio, and Ripple Effect Studios — shipped Battlefield 6 on October 10, 2025 for PlayStation 5, Windows, and Xbox Series X/S on a heavily rebuilt version of Frostbite, with Tactical Destruction and a new Kinesthetic Combat System as the headline features 1. The game sold over 7 million units in three days and became the biggest launch in series history 1.

The short answer: Frostbite is bad at being a general EA engine, but excellent at being a Battlefield engine. Battlefield 6 proves the difference.

Quick facts

Game: Battlefield 6, 18th entry in the Battlefield series, online-only first-person shooter 1
Release: October 10, 2025, PS5 / Windows / Xbox Series X/S; free-to-play RedSec battle royale on October 28, 2025 1
Engine: Frostbite, proprietary engine by DICE / EA, written in C++ and C# 2
Developer: Battlefield Studios — DICE leads multiplayer, Motive and Criterion lead campaign, Ripple Effect leads battle royale 1
Scale: 4 classes, 9 maps at launch, squads of 4, modes including Conquest, Breakthrough, Rush, Escalation, and Sabotage 1 3
Cost: Budget reported over $400 million, one of the most expensive games ever developed 1

What Frostbite actually is

Frostbite was created by DICE and is owned by Electronic Arts. It debuted in June 2008 in Battlefield: Bad Company and has since powered dozens of shooters and, controversially, RPGs, racing games, and action games 2. DICE itself was founded in 1992 in Sweden and has been an EA subsidiary since 2006; through Frostbite Labs it maintains the engine alongside Battlefield, Mirror's Edge, and Star Wars Battlefront 4.

The version history matters:

Frostbite 1 / 1.5: HDR Audio + Destruction 1.0 in Bad Company (2008), full building demolition in Bad Company 2 (2010) 2
Frostbite 2: Deferred rendering, real-time radiosity, and Destruction 3.0 in Battlefield 3 (2011) 2
Frostbite 3: Levolution in Battlefield 4 (2013) — dam destruction flooding the whole map — then expansion to Dragon Age: Inquisition, Mirror's Edge Catalyst, Battlefront, and Battlefield 1/V 2
Modern Frostbite: Rebuilt for Battlefield 2042 (2021) over 18 months, then again heavily rebuilt for Battlefield 6 with a new movement system and revamped destruction 2

EA briefly tried to make every studio use Frostbite. In December 2023 it reversed that strategy 2. That reversal is the key to understanding Battlefield 6.

Why Battlefield 6 stayed on Frostbite instead of Unreal

Three practical reasons, all visible on EA's own feature pages:

1. Destruction is gameplay, not decoration. EA markets Tactical Destruction as reactive, consistent, and tactical — buildings demolished with explosives, floors removed to open new paths 1 3. Unreal can do chaos physics, but Frostbite has 17 years of Battlefield-specific destruction tuning: HDR Audio that makes rubble sound different, network code that syncs collapsing cover, and map design built around Combat Zones for close-quarters inside large vehicle maps 1.

2. Combined arms needs one team owning the engine. Battlefield 6 mixes infantry, tanks in the Sahara, Gibraltar beach assaults, and New York defense in a single campaign plus large multiplayer 3. When DICE owns Frostbite, vehicle handling, audio occlusion, and server tick can be changed for Battlefield without waiting on Epic's roadmap. That ownership cost BioWare dearly, but it pays DICE back every map.

3. Institutional memory beats feature lists. Frostbite remembers what Battlefield feels like: suppressive fire, disabled-not-destroyed vehicles, class perks for specific weapons, revive-drag with the Kinesthetic Combat System 1. A general engine would need to relearn all of that. The $400M+ budget and four-studio Battlefield Studios structure only makes sense if the tech is already Battlefield-shaped 1.

After Battlefield 2042 was widely criticized, EA did not just change engines. It created Battlefield Labs, a closed worldwide playtest to collect feedback and tune the new game 1. That is how you de-risk a proprietary engine: keep the tech closed, open the feedback.

What Frostbite gained — and what it cost

Gained:

Believable scale without loading breaks, from infantry rooms to vehicle sprawl, with Tactical Destruction as the core differentiator 1 3
Signature audio-visual immersion: HDR Audio originated on Frostbite 1 and remains central to all-out warfare 2
Fast iteration for DICE on multiplayer, because the engine team and the game team share a building and a parent company 4

Cost:

Onboarding pain. Four studios, cultural clashes, burnout, and unrealistic EA sales expectations were reported during development 1. Proprietary tech amplifies coordination cost.
No easy hiring pool. Unreal developers are everywhere; Frostbite developers must be trained inside EA.
Past baggage. The same engine forced Need for Speed: The Run to spend a year re-tooling shooter tech for driving 2, and forced RPG teams to build dialogue and inventory systems from scratch.

Battlefield 6's open betas in August drew over 500,000 concurrent Steam players 1 — proof the cost paid off at launch, at least commercially.

The twist most coverage misses: Portal runs on Godot

Here is my favorite detail, and the most under-analyzed part of Battlefield 6's tech story.

The Battlefield Portal feature returned in Battlefield 6 with a complete overhaul. In 2042 Portal was a basic visual block editor. In Battlefield 6 it provides map creation and game-logic tools — and it is powered by the Godot engine 1.

Read that twice: the $400M Frostbite flagship outsources user-generated content to free open-source Godot.

Examples shown before launch included a recreation of Counter-Strike's Dust II with bomb defusal, plus zombies survival and racing modes. Within two days of launch the community had rebuilt Call of Duty 4's Shipment map and a Star Destroyer from Battlefield assets 1.

My take: this is humble and brilliant, and other proprietary-engine teams should copy it.

Frostbite is too hard and too licensed to give to players. Godot is easy, free, and already good at editors. Instead of dumbing down Frostbite or building a second editor from scratch, Ripple Effect and DICE let each engine do what it is good at — Frostbite for large-scale destruction simulation, Godot for accessible creation. It is a two-engine game wearing a one-engine marketing label.

This also reframes the old “Frostbite is bad at tools” complaint. Yes, Frostbite editor workflows famously frustrated non-DICE teams. Battlefield 6 does not claim to have fixed that for everyone. It routes around it: professionals suffer Frostbite, amateurs get Godot. That split is more honest than pretending one toolchain serves both.

My contrarian analysis: obscurity is the moat, destruction is the language

Most articles say Battlefield uses Frostbite because EA forces it to. I think causality now runs the other way: Battlefield needs an obscure engine.

If Battlefield 6 were on Unreal Engine 5, every design choice would be comparable to every other UE5 shooter — same Nanite rocks, same Lumen lighting language, same marketplace assets. On Frostbite, even a generic concrete wall feels proprietary because only Battlefield demolishes it that way. Destruction becomes a dialect only Frostbite speakers understand.

Three original ideas I have not seen elsewhere:

1. Walls are ammo. In most shooters cover is static. In Battlefield 6 on Frostbite, cover is a resource you spend. Blowing a floor to create a new path 1 is not vandalism, it is map editing under fire. That turns Frostbite's Destruction 1.0-to-4.0 lineage 2 into a combat economy: the team that budgets explosives better controls sightlines. Unreal-first shooters rarely think this way because their destruction is usually cosmetic.

2. Jank is the receipt. Frostbite games have a specific clumsiness — vehicles slide, debris kills you, revives drag bodies awkwardly with the Kinesthetic system 1 2. Players complain, then post clips. That awkward physics is proof the simulation is real, not animated. Clean UE5 movement would actually hurt Battlefield's marketing, because there would be fewer “you won't believe what happened” moments.

3. A closed engine needs open playtesting. Battlefield Labs 1 is not just marketing. When you own the engine, you cannot rely on thousands of external UE developers to find bugs. You must manufacture that community yourself. The 500k beta peak was load testing for both servers and the rebuilt Frostbite destruction pipeline. Proprietary tech shifts QA cost from license fees to playtest operations.

In short: Frostbite survives not because it renders better pixels, but because it remembers 17 years of Battlefield mistakes. That memory is not portable.

What indie devs should actually steal

Do not clone Frostbite. Steal the constraints:

Make one verb sacred. For Battlefield it is “demolish.” Pick one verb your engine does better than Unity/Unreal and build every map around it.
Split pro and community tools early. If your custom tech is hard, do what Battlefield 6 did: keep the hard core and ship creation in something friendly like Godot. Players do not care about engine purity.
Budget for feedback, not licenses. No engine fee still means paying for Labs-style playtests, beta servers, and iteration time. The $400M lesson is that proprietary savings are spent on people 1.
Record your feel. DICE's real asset is not C++ code, it is knowing how suppressive fire and vehicle disable should feel. Write that down. Small teams lose this when one programmer leaves.

Practice the Battlefield loop in your browser

You cannot run Battlefield 6's Frostbite destruction in a lightweight page, but you can drill the underlying skills — positioning, rotation, objective timing, and vehicle awareness — for free.

For close-quarters objective modes like Domination and Capture the Flag, Hazmob FPS: Online Shooter is a useful warm-up. Its seven-mode structure rewards the same habits Battlefield rewards: stick with a teammate in Team Deathmatch, hold zones instead of chasing kills in Domination, and save utility for plants and choke pushes.

If you struggle with high-ground takes and endgame rotation — the Zone Wars problem that also decides Escalation funnels in Battlefield 6 — BuildNow GG trains exactly that. Its V-Arena duels and Zone Wars shrinking zones force early rotation and box-peek discipline that transfers to holding Combat Zones.

For the combined-arms half — driving to hotspots, hijacking closer rides, ramming groups on wide roads then clearing on foot — Sandbox City - Cars, Zombies, Ragdolls is the closest browser analogue. Practice the routine it teaches: drive to the marker with E to enter and exit, park before looting, and use slow motion to line up tricky shots. Battlefield never has zombies, but the drive-park-clear loop is identical.

Broader inspiration, including vehicle shooters and battle-royale variants, is collected on EasyHub's Battlefield 6 topic page, which frames browser war games as lightweight alternatives when you do not have 100 GB and an RTX-class PC.

Bottom line

Battlefield 6 bet on Frostbite because destruction, combined arms, and 17 years of Battlefield-specific tuning outweigh the hiring and tooling pain — and then cleverly dodged the tooling pain for creators by building Portal on Godot.

That hybrid — closed simulation core, open creation shell — is the real lesson. The future of niche engines is not beating Unreal at everything. It is knowing what to keep proprietary and what to give away.

Why Kingdom Come: Deliverance II Bet on CryEngine: The Outdated Shooter Tech Behind a 6-Million-Copy Medieval Epic

dev game — Wed, 21 Oct 2026 20:34:44 +0000

Kingdom Come: Deliverance II should not have worked on paper. A 250-person studio in Prague shipped a first-person, open-world medieval RPG with two large maps, layered armor, horses, and a crime system that remembers what you stole — on CryEngine, a German shooter engine from 2002 that most players associate with Far Cry and Crysis, not dialogue trees.

Warhorse Studios did it anyway. The game released for Windows, PlayStation 5, and Xbox Series X/S on 4 February 2025, uses CryEngine as its credited engine, and sold more than 6 million copies by June 2026. The first game, also built on a modified version of CryEngine, had already sold over 10 million units by May 2025 after launching on 13 February 2018.

My take, stated upfront and separated from the verified facts below: staying on CryEngine was not nostalgia. It was a daylight-and-debt decision. Warhorse traded hiring ease and smooth launch for forests that look right at noon, armor that behaves like clothing, and a team that already knew where the bodies were buried in the code.

The short answer

Warhorse stayed on CryEngine because KCD2 is a daylight forest sim with swords, not a dungeon crawler with magic. CryEngine was built for long view distances, dense vegetation, natural daylight, and streaming open terrain — the same tech that powered Far Cry jungles and Crysis islands. For a 1403 Bohemia made of fields, woods, Kutná Hora streets, and castles seen from horseback, that fit was closer than Unreal Engine 5's headline features.

The studio also carried seven years of custom work forward: first-person melee with directional attacks and blocks, horse systems, lockpicking and pocket-picking minigames, reputation and crime punishment including fines, caning, branding, and the pillory, plus fast-travel encounters and a hardcore mode without compass or location markers. Rebuilding all of that on a new engine would have cost the “twice as large” map that defined the sequel.

Facts first: what the game and the engine are

Warhorse Studios was founded on 25 July 2011 in Prague by Dan Vávra and Martin Klíma, veterans of 2K Czech and Bohemia Interactive. It employed about 250 people in 2024, was acquired by Koch Media, now Plaion, in February 2019 for €42.8 million, and announced KCD2 in April 2024 as twice as big as the original.

KCD2 is a single-player, first-person, open-world action RPG published by Deep Silver and directed by Daniel Vávra. Unlike the first game’s single connected world, the world is twice as large and split into two larger freely explorable areas. Melee options include short swords, longswords, maces, axes, hammers, and polearms, some usable with a shield, with directional attacks, blocks, combos, and counterattacks, some unlocked through quests. Ranged options are bows, crossbows, and early firearms such as the hand cannon. You can fight on horseback, but some ranged weapons must be loaded before mounting and cannot be reloaded while mounted.

The crime layer is unusually strict: lockpicking, pocket-picking, murder, assault, and poaching can be spotted or deduced by NPCs, leading to arrest, fighting or fleeing, fines, caning, branding, time in the pillory, or talking your way out with dialogue skills. Stolen goods are marked with a red hand in inventory. Fast travel uses a compass and map with a choice to engage or avoid encounters at a shown percentage chance. A post-release hardcore mode removes fast travel, hides compass direction, player location, health and stamina bars, and adds starting debuffs.

CryEngine itself is designed by German developer Crytek, first released on 2 May 2002, with stable release 5.7.1 on 19 May 2022, written in C++, with Qt, Lua, and C#. It supports Linux, Windows, Nintendo Switch and Switch 2, PlayStation 3 through 5, Wii, Wii U, Xbox 360, Xbox One, and Xbox Series X/S under a source-available commercial license with a royalty model. Crytek describes it as an AAA all-in-one 3D solution proven on Hunt: Showdown, Crysis, and The Climb, with full source code and all features.

Its history explains why it feels unfashionable: it powered all Crytek titles starting with Far Cry, was used for Sniper: Ghost Warrior 2 and SNOW, Warhorse uses a modified version for Kingdom Come, Ubisoft maintains a heavily modified fork called Dunia from the original Far Cry for later Far Cry games and The Crew 2, and it was licensed to Amazon in 2015 for a reported $50–70 million, becoming Amazon Lumberyard in February 2016 and later the basis for Open 3D Engine. Core strengths listed for the engine include simultaneous WYSIWYG Sandbox editing, vegetation and terrain cover generation, road and river tools, flexible time-of-day, streaming, and performance analysis tools, plus natural lighting, dynamic soft shadows, deferred lighting, real-time global illumination, and volumetric fog and light shafts.

What staying on CryEngine gained

1. Daylight that sells history. Most RPG engines show off in torchlight or neon. KCD2 has to sell 2 p.m. in a wheat field. CryEngine’s time-of-day, irradiance volumes, eye adaptation, and vegetation cover tools were made for exactly that: sun through leaves, haze over a valley, mud that looks wet without ray-traced magic. That is why screenshots read as “photo of Bohemia” rather than “fantasy.”

2. A double-size map with the same Prague team. KCD2 did not double headcount to double the map. It reused quest logic, horse logic, clothing layers, and AI routines tuned since 2018. Warhorse’s own history shows the payoff: the first game reached 6 million units by February 2024, and the sequel sold 1 million copies in a single day and 4 million by November. You do not throw away that tuning to chase a logo.

3. Weight that fits 1403. Directional swordplay, stamina drains, cannot-reload-on-horse rules, and a save system that punishes theft are friction. On a smooth new engine the temptation would be to smooth them away. On CryEngine they stayed clunky in a way that matches a blacksmith’s son who is bad with a sword for the first 20 hours.

What it cost

The cost is visible and should not be minimized. The first game was praised for story but criticized for technical bugs and its save system, a reputation the sequel inherited at launch. Hiring is harder: Prague graduates learn Unreal and Unity, not a modified CryEngine with Lua and Flow Graph history. Porting is slower too: the original took until March 2024 to reach Nintendo Switch, and KCD2’s Switch 2 version is scheduled for 2027, two years after the main launch.

That delay is the tax for a custom branch. When Crytek moves to 5.7 and beyond, Warhorse must backport or ignore fixes. When Sony and Microsoft update SDKs, a 250-person team cannot rely on thousands of Unreal forum answers.

My analysis: three ideas you will not find in the press releases

These are interpretations, not verified facts. I label them as such to keep facts and opinion separate.

Idea 1: CryEngine is a forest engine, Unreal 5 is a rock engine. Everyone talks about Nanite geometry and Lumen bounce light. KCD2’s hardest problem is not a 10-million-triangle cathedral. It is 10,000 trees that sway, cast soft shadows at noon, and occlude an ambush 200 meters away without popping. CryEngine’s vegetation, view-distance fogging, and streaming were built for that jungle problem since Far Cry. My contrarian claim: for low-verticality, high-foliage, daylight realism, an “old” vegetation pipeline beats a new geometry pipeline. Warhorse did not need better rocks. It needed better wheat.

Idea 2: Jank is a realism device, not just a bug. Most reviews treat KCD’s clumsy grapples and fussy lockpicks as failures to be patched. I think they function like accent in a historical film. When opening a door takes two hands and picking a pocket risks branding, you stop playing like an assassin and start playing like a peasant who is afraid of guards. A frictionless engine would have made Henry feel like Spider-Man in a gambeson. The friction teaches class.

Idea 3: Obscurity is a hiring moat. The common advice is “use Unreal so you can hire.” Warhorse inverted it: by staying on a modified CryEngine, it kept a team that cannot be easily poached by every Unreal startup in Europe, and it made its game harder to clone. Anyone can buy Quixel medieval assets for Unreal. Almost no one has seven years of tuned plate-armor layering, horse-mounted archery limits, and camp AI on CryEngine. Boring tech, in this case, is copy protection.

If you came here asking “is KCD2 Unreal Engine 5?” — you are not alone. Search suggestions show that exact confusion, with variants like “kingdom come deliverance 2 unreal engine 5” and “what engine is kingdom come deliverance 2” appearing alongside cheat-engine queries. The answer is no: credited engine is CryEngine, and the UE5 look comes from art direction and daylight, not from Nanite.

What indie developers can steal

You do not need CryEngine to use the lesson. Use this checklist:

Pick the engine that already solves your hardest light. Forest noon? Snow night? Neon rain? Match the engine’s origin biome to yours.
Price a switch in content, not licenses. Count how many weapons, quests, and AI behaviors you would have to retune. For Warhorse that number was larger than any royalty.
Keep one intentional friction. If your fantasy is “be a nobody in 1403,” keep lockpicking hard, keep horses scary, keep theft marked with a red hand.
Budget for the tax. Obscure tech means slower ports and fewer tutorials. Plan for a 2027-style late port if you branch.
Exploit obscurity. A small, stable team that knows its tools beats a large team relearning a popular engine mid-project.

Train the feel in a browser

You cannot practice longsword half-swording in a tab, but you can train the three skills KCD2 actually tests: horse control, directional timing, and patience under watch.

For horse movement and long routes, try Horse Simulator 3D. Its island forces you to manage strength, speed, and endurance across forests, rivers, and hills while avoiding wolves and finishing races — a low-stakes way to feel why Warhorse forbids reloading on horseback and why stamina matters on a long ride.

For melee timing without combos to memorize, try Iron Snout. This SnoutUp arcade brawler throws wolves from both sides and asks for centered positioning, directional punches and kicks, catching thrown axes, and not spamming. That discipline — observe, then strike left or right — is the same loop as KCD2’s directional attacks and blocks.

For crime patience, try Murder. You hold Space to raise the knife only while the king looks away, watch for coughs and twitches, and get sent to the dungeon if you hold through a tell. It compresses KCD2’s pickpocket-and-pillory anxiety into 30-second loops: build meter, release, wait.

None of these are medieval sims, and none run on CryEngine. They are useful precisely because they isolate one verb — ride, block, wait — that KCD2 refuses to automate.

Limits and next questions

This account relies on credited engine listings, release and sales milestones, and Crytek’s feature descriptions, all accessed in October 2026. It does not have Warhorse’s internal profiler data, console frame-rate breakdowns, or the exact royalty terms versus Unreal Engine 5’s 5% after $1 million. Those would change the cost math but not the design point: Warhorse chose fit and continuity over fashion.

Open questions worth watching: will the 2027 Switch 2 port require cutting vegetation density that defines the game’s look? Will the two announced open-world RPGs — another Kingdom Come adventure and a Middle-earth Lord of the Rings project — stay on the same modified CryEngine, or will team growth force a switch? If they stay, my moat idea gets stronger. If they leave, the hiring tax finally outweighed the forest dividend.

Why Elden Ring Bet on a Nameless In-House Engine: The Old Souls Code Behind a 30-Million-Copy Open World

dev game — Wed, 21 Oct 2026 20:05:13 +0000

Elden Ring did not bet on Unreal, Unity, or any licensed engine with a logo splash. It bet on FromSoftware's own nameless, internally evolved Souls codebase — the same lineage that shipped Demon's Souls, Dark Souls III, and Sekiro — stretched for the first time across a true open world.

That choice looks stubborn until you count the payoff: over 30 million copies sold, Game of the Year wins, and a 10-million-unit expansion, all from a Tokyo studio that listed 456 people in May 2025. Elden Ring was developed by FromSoftware and published by Bandai Namco Entertainment, released February 25, 2022 for PS4, PS5, Windows, Xbox One and Series X/S, with a Nintendo Switch 2 version on August 28, 2026. It has sold over 30 million copies and is cited among the greatest games of all time. My contrarian read: the old engine was not baggage. It was the design document.

What the bet actually was

There is no official engine name to license, no asset store, and no Epic or Unity credit in the official materials. Bandai Namco's official Elden Ring site credits only FromSoftware, lists action RPG / open world, single-player plus online up to 3 players and up to 6 in Arenas, and never mentions a third-party engine.

What we can verify is continuity:

Same lead programmer, same publisher Bandai Namco Entertainment, same director Hidetaka Miyazaki. That is not proof of identical code, but it is strong evidence of an internal iteration rather than a rip-and-replace to Unreal.

It was also not Sony's shared PhyreEngine. PhyreEngine is Sony's free-to-use proprietary engine used in over 200 titles, and its public game list does not include FromSoftware's Souls titles. FromSoftware, founded in 1986 as business software and shifted to games with King's Field in 1994, kept its tech in-house while peers standardized on commercial engines.

In search terms this creates a revealing confusion. English suggestions in October 2026 cluster around elden ring engine, was elden ring made with unreal engine, and what game engine is elden ring — alongside polluting cheat engine and mod engine results. Players assume a 30-million seller must be Unreal. The fact they have to ask is the niche: a blockbuster with no engine brand.

What the old engine bought

1. Combat that refuses to be cancelled

Soulslike combat is methodical: stamina management plus animation-priority actions that prevent cancelling movement until the animation plays out, leaving you vulnerable. That stiffness is usually criticized as clunky. In Elden Ring it is the product.

Unreal's character templates want responsive, blendable cancels. From's engine enforces commitment: a colossal sword swing, a flask sip, a Torrent dismount all lock you in. Bosses are tuned around that lock-in. Change the engine and you would have to retune 100+ bosses or betray a decade of muscle memory. Keeping the engine kept the timing contract intact while the map grew 10x.

2. A small team shipping an open world

FromSoftware lists King's Field, Armored Core, Demon's Souls, Dark Souls, Bloodborne, Sekiro and Elden Ring as its core products, with 456 employees in May 2025. Ubisoft-style open worlds need thousands. From did it with hundreds because it reused systems: bonfire logic became Sites of Grace, souls became runes, covenants became summon signs and invasions.

Elden Ring keeps Souls loops — Sites of Grace for leveling and fast travel, runes dropped on death and lost forever if you die again, Spirit Ash summons near Rebirth Monuments. Soulslikes typically use bonfire-like checkpoints for respawn, leveling and fast travel while resetting enemies. From did not rebuild open-world systems. It reskinned proven ones and spent its budget on placement, not plumbing.

3. Cross-generation shipping without a rewrite

The same build spans PS4 and PS5, Xbox One and Series X/S, Windows, then Switch 2 four years later. A commercial engine would have made that easier on paper, but also forced a mid-project version migration. From's frozen, well-understood codebase let it ship where players actually were in 2022 — including 100-million-plus PS4s — instead of chasing features.

What it cost, honestly

The bill came due on PC. Elden Ring launched on DirectX 12 with infamous shader-compilation stutter, frame-pacing issues, and no day-one ultrawide or ray-tracing showcase. Ray tracing only arrived later as a limited patch, and mod support remains unofficial and fragile compared to Unity or Unreal mod ecosystems.

Shadow of the Erdtree, released June 21, 2024, sold ten million units by July 2025 despite renewed complaints about difficulty spikes and performance on dense areas. Nightreign, the roguelike co-op spin-off released May 30, 2025, topped May sales charts and sold over 5 million units — again on familiar tech, not a new renderer.

That is the trade: reach and feel over smoothness and tooling. It worked commercially, but it is not free.

Three ideas you will not find in the wiki summary

These are my analysis, not FromSoftware statements. I label them as such because the studio never published an engine postmortem.

1. Torrent is a streaming budget wearing a horse costume.

Torrent is usually praised as traversal fun. I think Torrent's speed, double-jump height, and the spacing of Sites of Grace are engine budgets made visible.

From's engine was built for interconnected corridors with elevator-mask loading, not 6 open areas you can gallop across. The game has six main areas you roam freely, with Torrent as the main mode of transport plus fast travel. My read: Torrent gallops just fast enough to feel free, just slow enough that a PS4 HDD can stream the next field, and Sites of Grace are placed where the streamer needs a hitch. Stakes of Marika near bosses are not kindness — they are respawn points that avoid reloading the whole field. If From had used Unreal World Partition, Torrent could be twice as fast. It would also stop being scary, because speed kills dread.

Practice the idea cheaply in a browser: Horse Simulator 3D gives you an island, stamina management, and family progression. It is not Elden Ring, but roaming, judging gallop distance, and deciding when to sprint versus conserve maps surprisingly well to Torrent routing between Graces.

2. Stutter preserved fairness.

Most coverage treats stutter as pure failure. I think it accidentally protected the game's core promise: identical i-frame timing for everyone.

Unreal and Unity push precompiled shaders, dynamic resolution, and frame interpolation that vary timing frame to frame. Souls dodge rolls live or die on 2-3 frame windows. From's older, more deterministic pipeline stuttered badly on new drivers, but when it ran, it ran the same roll every time. Players hated the hitches and still trusted the parry. That trust is why Soulslike death loops — lose currency, retrieve it, lose it forever on a second death feel fair in Elden Ring and cheap elsewhere. Smoothness without determinism would have broken the genre.

If you want to feel that discipline without a 100-hour commitment, try Iron Snout. Its rules are brutally Souls-like: stay centered, watch patterns, do not spam attacks. The game punishes button mashing the same way animation priority does — one mistimed kick ends the run.

3. Having no engine name is a hiring and hype moat.

Unreal developers are portable. From developers are not. By never branding the engine, FromSoftware avoids a public API, public roadmap, and public comparison. Job posts ask for Souls experience, not certificate keywords. Competitors cannot poach a team by offering a newer version of the same tool. Fans cannot demand Nanite or Lumen because there is no toggle to demand.

It also lowers expectations in a useful way. When a UE5 game hitches, players call it unoptimized. When Elden Ring hitches, players call it FromSoftware. The studio converted a tech deficit into auteur identity — the jank is part of the lore, like Mimics and messages on the floor.

What indie devs can steal without 30 million sales

Freeze your tech one project early. From iterated the same combat, netcode for up to 3-player co-op, and checkpoint logic across five games. Your prototype's janky controller is fine if you keep it for three releases and tune content to it.
Make mounts and checkpoints do streaming work. If your open field hitches, do not buy a new engine. Add a horse that is 15% slower than fun, and place respawn points to hide reloads. Players will praise the horse.
Keep animation commitment. Let attacks lock the player in. It feels bad for ten minutes, then feels weighty forever. Modern blend-everything controllers feel good for ten minutes, then feel weightless forever.
Ship where players are, not where engines demo. PS4 + Xbox One support in 2022 looked backward. It paid for the DLC.

For dragon encounters specifically — Elden Ring's flying dragons punish ground greed and reward hit-and-run breath management — a low-stakes trainer helps: Dragon Simulator 3D teaches the same loop of starting from the air, softening with breath, then finishing close, plus managing health, damage and speed upgrades. Think of it as Agheel with training wheels.

Limits and open questions

FromSoftware has not disclosed source code, language, renderer details, or internal codename, so any claim that Elden Ring reuses exactly which Dark Souls III modules remains inference from credits and behavior, not confirmed fact. Sales figures above are as reported on Wikipedia in October 2026 and will keep moving, especially with the Switch 2 Tarnished Edition bundling base game plus expansion.

The film adaptation slated for 2028 and future balance patches could also change canon, but they will not change the engine lesson: control your timing, control your streaming, and you can ship an open world without renting someone else's future.

Why Apex Legends Bet on Source: The 2004 Engine Behind a 100-Million-Player Battle Royale

dev game — Wed, 21 Oct 2026 08:35:07 +0000

Why Apex Legends Bet on Source: The 2004 Engine Behind a 100-Million-Player Battle Royale

Short answer: Apex Legends runs on a heavily modified version of Valve's Source engine, not Unreal or Unity. Developed by Respawn Entertainment and published by Electronic Arts, it launched February 4, 2019 as a free-to-play battle-royale hero shooter set in the Titanfall universe. Apex Legends - Wikipedia The mobile version later used Unreal Engine 4, but PC and console Apex is Source all the way down.

That sounds absurd in 2019. Fortnite was on Unreal Engine 4, PUBG was on Unreal Engine 4, Call of Duty: Warzone would be on IW Engine. Respawn shipped a 60-player live-service hit on tech that debuted with Half-Life 2 in 2004.

What engine does Apex actually use?

Engine: Source (Valve, C++, Havok middleware), heavily modified. Listed as Source on PC/console, with Unreal Engine 4 for mobile. Apex Legends - Wikipedia
Lineage: Titanfall (March 11, 2014) ran on Source. Titanfall (video game) - Wikipedia Titanfall 2 (October 28, 2016) ran on a heavily modified version of Valve's Source. Titanfall 2 - Wikipedia Apex inherited that fork, then modified it again for 60 players, Legends abilities, and a single large map instead of 6v6 arenas.
Source itself: 3D engine by Valve, successor to GoldSrc in 2004 with Half-Life: Source, Counter-Strike: Source and Half-Life 2. Used for Team Fortress 2, CS:GO, Dota 2, Portal and Left 4 Dead, plus third-party games including most Titanfall games, Vampire: Bloodlines, Dear Esther and Garry's Mod. Written in C++, source-available SDK, succeeded by Source 2 in 2015. Source (game engine) - Wikipedia
Who built it: Respawn was founded April 2010 by Jason West and Vince Zampella after they were fired from Infinity Ward, where they led Call of Duty 4: Modern Warfare. By July 2010, 38 of 46 Infinity Ward resignations had followed them to Respawn. EA acquired Respawn in December 2017 for $151M cash plus equity. Apex debuted with no prior marketing via YouTube livestream February 2019. Respawn Entertainment - Wikipedia
Scale: Free-to-play, games-as-a-service, cross-platform multiplayer. 3-player squads, 20 squads per battle-royale match. 25 million players by end of first week, 50 million in first month, approximately 100 million by April 2021, making it one of most-played games by player count. Apex Legends - Wikipedia Minimum PC spec is still modest: 64-bit Windows 10, 6 GB RAM, AMD HD 7790 / GTX 950, 75 GB storage. Apex Legends on Steam

In other words, Apex is Titanfall without Titans, without wall-running, with twice the players, on an engine that was already 15 years old at launch.

Why not Unreal or a new in-house engine?

Everyone assumes battle royale = Unreal. Respawn had three practical reasons to refuse:

1. The team knew Source in their hands, not in a manual. Respawn's core came from Modern Warfare, not from Epic's ecosystem. By 2011-2014 they had already bent Source to do what Source was never meant to do: wall-running Pilots, double-jumps, Titans dropping from orbit, 50 active characters with AI grunts offloaded to cloud. Titanfall (video game) - Wikipedia Throwing that away in 2016-2018 for Unreal would have meant re-learning animation, gun tuning and server prediction while EA expected a live-service hit. Familiarity beat features.

2. Gunfeel is the product. PC Gamer's launch review nails it: guns have ballistics, drop and distance matter in the same split-second as aim, auto-equip picks the better attachment so you are not menuing in PUBG hell. Apex Legends review - PC Gamer That peppy Modern Warfare snap — Wingman hand-cannon, Peacekeeper star-pattern pellets, Triple Take horizontal spread — came from a decade of Source recoil and hitscan tuning Respawn owned. Unreal would have given better foliage. Source gave predictable bullets.

3. Low spec is distribution. HD 7790 / GTX 950 minimum keeps internet cafes, school PCs and last-gen consoles in queue. Apex Legends on Steam Like my read on Counter-Strike 2's toaster strategy, accessibility is matchmaking. A 100-million-player pool forgives more than a ray-traced puddle ever will.

My take — and this is analysis, not Respawn PR: Respawn treats engines like a veteran band treats a beat-up guitar. While rivals shop for tonewoods, they kept the instrument that already knows their hands. The boring fork is a moat.

What Source actually gave Apex

Forget marketing specs. Three Source traits shaped how Apex feels:

1. Slide, not wall-run: speed without vertical chaos

Titanfall 2 chained wall-runs, double-jumps, slides, pulse blades and grapples. Titanfall 2 - Wikipedia Apex cut Titans and wall-running outright, but kept slide-on-slope, climb-over-short-walls, and ziplines. Apex Legends - Wikipedia PC Gamer called sliding both a skill ceiling and a defensive posture: hit crouch downhill to toboggan, chain staircases for free speed, shrink your hitbox under fire. Apex Legends review - PC Gamer

On Source's brush-based maps with readable collision, slide is cheap to predict and cheap to render. No need for Unreal Chaos physics. You get Titanfall momentum with battle-royale readability: one plane to watch, not six walls.

2. Ping: voice chat you do not have to hear

Apex's nonverbal ping system lets you mark directions, weapons, enemies and strategies without speaking. Apex Legends - Wikipedia The review describes object-sensitive depth: ping an enemy and the marker turns red, ping an open door and your Legend says someone has been there, ping a scope and the character calls out that exact scope, with different lines for open vs closed loot. Apex Legends review - PC Gamer

This is engine pragmatism disguised as accessibility. Source's entity system already knew what every item was. Exposing that to a single middle-mouse button turned strangers into squads in silence — critical when you queue 20 random trios who will never join Discord.

3. Kings Canyon readability over photorealism

Apex's island mixes cliffs, swamps, Mad Max arenas and giant skeletons, but no single area is a visual masterpiece — diversity helps you orient while migrating between regions, with balloons and supply ships scattering players to avoid Tilted Tower crowding. Apex Legends review - PC Gamer

That is Source 2004 logic: baked lighting, tight corridors, strong silhouettes. Legends have chunky hitboxes and loud abilities — Gibraltar dome, Wraith void, Caustic gas — because the engine cannot sell photoreal faces at 60 meters. So it sells readable chaos instead. You always know who shot you.

What it cost: tick rate, content pain, and a split codebase

Staying on Source was not free. Three costs still follow Apex:

The 20 Hz argument never dies. Respawn never marketed sub-tick like Valve did for CS2. Players feel hit-reg and no-reg clips more at 60 players than at 5v5, and every season revives engine-update threads. My read: Source's server model was built for 32-player Counter-Strike, not 60-player hero abilities with ziplines and balloons. Respawn papered it with prediction and small maps, but cannot delete the math.
Content is slower than Unreal's pipeline. No Nanite, no Lumen, no Mega sunlight slider. Every Legend, heirloom and Town Takeover must fit a hand-rolled renderer and old tool chain. That is why maps evolve by Takeovers and Mixtape modes (Team Deathmatch 6v6, Control 9v9, Gun Run) rather than fresh islands every season. Apex Legends - Wikipedia
Mobile had to defect. The infobox split — Source on PC/console, Unreal Engine 4 on mobile — is an admission. Touchscreen, variable GPUs and 2022 mobile live-ops did not fit the fork. The mobile version was discontinued in 2023 anyway, which tells you how hard split-engine live service is. Apex Legends - Wikipedia

Switch ports (March 2021), PS5/Xbox Series upgrades (March 2022) and Switch 2 (August 2025) all had to be coaxed from the same base. Apex Legends - Wikipedia Unreal would have made ports trivial. Source made them handcraft.

Three ideas most coverage misses

These are my interpretations, not Respawn statements. Take them as lenses, not facts.

1. Old engine as anti-bloat filter. Every modern battle royale adds vehicles, building, pets, weather. Apex could not afford to, because Source would choke. So it added Legends — cooldowns and passives that reuse existing physics. Gibraltar's dome is a sphere trigger. Pathfinder's zipline is Titanfall code with a new skin. Constraint forced design into characters, not systems. I think Apex survived where flashy new-tech rivals died because Respawn had no budget for gimmicks. Poverty of engine became richness of roster.

2. The wall-run apology. Cutting wall-running looks like a downgrade. I think it was a confession: wall-running is brilliant at 6v6 and unreadable at 60 players. Imagine trying to ping a wall-running Wraith across three buildings. Slide keeps 80% of the joy (free speed, downhill chains, dodge posture) with 20% of the chaos. Most outlets frame this as loss. I frame it as editing. Respawn killed their signature to save readability — the hardest cut in sequel design.

3. Familiarity as onboarding. COD veterans who never touched Titanfall pick up Apex and instantly strafe, slide-cancel and thirst pings. That is not accident. Respawn founders made Modern Warfare, then made Titanfall feel like Modern Warfare with jetpacks, then made Apex feel like Titanfall without jetpacks. Each step sheds one verb. An Unreal reboot would have reset muscle memory. The Source fork preserved it. In a genre where first-10-minutes retention decides everything, feeling home in 30 seconds beats looking next-gen in trailers.

What indie devs can steal (without EA's budget)

Keep the gun, change the game. If your prototype already shoots well, do not re-platform for graphics. Respawn kept Source gunfeel from Titanfall to Apex and changed everything around it: player count, objectives, characters. Players forgive old pixels. They never forgive new recoil.
One button for teamwork. If randoms cannot cooperate without voice, you have a UI bug, not a community bug. Apex ping works because Source already named every entity. Name everything in your game too, then let one key speak that name aloud.
Cut your signature. Your coolest mechanic may not survive scale. Prototype at full player count early, then kill what becomes noise. Slide over wall-run at 60 players is a template: keep the fantasy (speed), drop the cost (vertical tracking hell).
Spec for cafes, not trailers. Target GTX 950 plus 6 GB plus 75 GB before you target RTX. Apex Legends on Steam A bigger pool shortens queues, dilutes smurfs, and funds seasons longer than any lighting model.
Fork, do not restart. Heavily modified Source powered Titanfall, Titanfall 2 and Apex across 2014-2019+. Titanfall (video game) - Wikipedia Titanfall 2 - Wikipedia Each game paid down tech debt for the next. If you own a small engine, version it like a savings account, not a rewrite.

Feel the ideas in a browser (no 75 GB download)

You cannot practice Apex Legends recoil without Apex, but you can train the underlying skills Source rewards: rotation timing, close-quarters aim, and ring psychology.

For rotation and endgame — high-ground takes, early rotates, boxing up in final circles — warm up in BuildNow GG's Zone Wars. Its own guide frames Zone Wars as 6-player shrinking-zone practice for rotation and high-ground retakes, with Freebuild for 90s and edits first. That loop mirrors Apex third-party avoidance: rotate early, use one wall and ramp for crossings, save utility for the last ring.
For raw FPS duels and team roles — trading kills, holding angles, playing objectives vs chasing kills — try Hazmob FPS: Online Shooter. Its seven modes (Team Deathmatch for beginners, Domination for objective holding, Search and Destroy with no respawns per round) teach the same lesson Apex pings teach: winning fights and winning the match are not the same.
For last-squad psychology — when to hide, when to third-party, how shrinking space forces mistakes — mess with Squid Game Online. It is a 30-player 7-round survival game, not a shooter, but its core pressure (one mistake ends the run, Glass Bridge punishes rushing, Judgment Night forces a brawl when too many survive) is a playful mirror of Apex ring logic: patience beats greed until the circle says otherwise.

None of these replace slide-jumps or Legend cooldowns. They let you rehearse decisions — rotate, trade, survive — in a tab while Apex patches.

FAQ

What engine does Apex Legends use?
A heavily modified version of Valve's Source on PC and console, with Unreal Engine 4 used for the now-discontinued mobile version. Apex Legends - Wikipedia

Is Apex in the Titanfall universe?
Yes. Set in the same science-fiction universe as Respawn's Titanfall series, with Titanfall characters appearing as Legends. Apex Legends - Wikipedia

When did Apex Legends release?
February 4, 2019 for PS4, Windows and Xbox One, with no prior marketing — a surprise shadow-drop. Switch in March 2021, PS5/Xbox Series in March 2022. Apex Legends - Wikipedia

How many people play Apex?
25 million by end of first week, 50 million in first month, about 100 million by April 2021. Apex Legends - Wikipedia

What engine did Titanfall use?
Source for Titanfall (2014) and a heavily modified Source for Titanfall 2 (2016). Titanfall (video game) - Wikipedia Titanfall 2 - Wikipedia

Why does Apex feel like Call of Duty?
Respawn was founded by the Infinity Ward leads behind Call of Duty 4: Modern Warfare, who brought hitscan tuning and movement philosophy with them. Respawn Entertainment - Wikipedia

My bottom line: Apex Legends is not impressive despite running on a 2004 engine. It is impressive because Respawn refused to leave it. The Source fork forced readable fights, cafe-friendly specs, and character design over system bloat — then turned slide and ping into a battle royale language. In a market addicted to new engines, Respawn won by staying in the museum and learning every echo.

Why Counter-Strike 2 Bet on Source 2: The Boring In-House Engine Behind Tactical FPS's Smartest Upgrade

dev game — Wed, 21 Oct 2026 08:05:05 +0000

Why Counter-Strike 2 Bet on Source 2: The Boring In-House Engine Behind Tactical FPS's Least Exciting — and Smartest — Upgrade

Short answer: Counter-Strike 2 runs on Valve's proprietary Source 2 engine, not Unreal or Unity. Announced on March 22, 2023 and released on September 27, 2023 as a free upgrade that replaced CS:GO on Steam, it kept the same 5v5 bomb-defusal core and bet everything on invisible tech: sub-tick networking, dynamic volumetric smoke, and physically-based lighting. Counter-Strike 2 - Wikipedia

That sounds dull next to Unreal Engine 5 demos. It is the point. Valve waited 11 years since CS:GO (August 21, 2012) to ship a sequel that looks like a patch, because the real product was never graphics — it was trust in every bullet registration.

What engine does CS2 actually use?

Engine: Source 2, developed by Valve, written in C++, proprietary license. Source 2 - Wikipedia
Predecessor: CS:GO ran on original Source. Counter-Strike 2 moved to Source 2 with improved graphics and a new sub-tick server architecture. Counter-Strike 2 - Wikipedia
How you get it: Steam describes it as a free upgrade to CS:GO and the largest technical leap in Counter-Strike's history, built on Source 2 and modernized with realistic physically-based rendering, state-of-the-art networking, and upgraded Community Workshop tools. Counter-Strike 2 on Steam
Lineage: The series began in 1999 as a Half-Life mod by Minh Le and Jess Cliffe before Valve acquired it and released retail Counter-Strike in 2000. Counter-Strike: Source was the first entry on the original Source engine, CS:GO stayed on Source, CS2 is the first Counter-Strike on Source 2. Counter-Strike - Wikipedia

Source 2 itself was announced at GDC in March 2015 as the successor to Source, with the intent to allow content to be created more efficiently, support the Vulkan graphics API, and use a new in-house physics engine called Rubikon to replace Havok. The first game to use it was Dota 2, ported from Source in the Dota 2 Reborn update in September 2015. Other Source 2 games include Artifact, Dota Underlords, Half-Life: Alyx, Counter-Strike 2, and Deadlock in beta. Source 2 - Wikipedia

Valve president Gabe Newell said the company prioritized its own games before releasing the SDK publicly, intending to make the engine free for developers as long as the game is published on Steam. Source 2 - Wikipedia

Why not Unreal Engine 5 or Unity?

Everyone asks this. On paper, Unreal would have given Valve Nanite, Lumen, and a hiring pipeline. In practice, three constraints made staying in-house rational:

1. Inventory is the engine. Steam promises all items from CS:GO moving forward to CS2. Counter-Strike 2 on Steam Ten years of skins, stickers, cases, and market listings could not be re-imported into a licensed engine without re-validating materials, wear ratings, and lighting. A clean Unreal rewrite would have risked the one thing Valve cannot re-ship: player trust that a rare knife still looks and trades the same.

2. Networking is the product. CS lives or dies on how a peek feels at 1280x960 on a 2014 i5. Source 2 let Valve keep deterministic recoil patterns, community servers, Kreedz climbing physics quirks, and Valve Anti-Cheat in one stack. CS:GO matchmaking ran on dedicated Valve servers plus heavily moddable community servers. Counter-Strike: Global Offensive - Wikipedia Moving to Unreal would have meant re-tuning a decade of muscle memory.

3. SteamOS is the strategy. Source 2 supports Vulkan, and CS2 ships with explicit Linux requirements: Ubuntu 20.04, AMD GCN+ or NVIDIA Kepler+ with up-to-date Vulkan drivers, VK_EXT_graphics_pipeline_library recommended. Minimum Windows spec is strikingly low: i5 750, 8 GB RAM, 1 GB DX11 card, 85 GB storage. Counter-Strike 2 on Steam That is not laziness. That is distribution: internet cafes, school PCs, and Linux machines stay in the pool.

My take — and this is analysis, not Valve PR: Valve treats engines like plumbing. While rivals sell engines as trailers, Valve sells continuity. The boring choice is a moat.

What Source 2 actually changed in CS2

Forget photorealism. The three changes that matter are all about fairness and readability.

1. Sub-tick: killing the tick-rate debate

CS:GO's 64-tick vs 128-tick argument dominated forums for a decade. Community 128-tick servers felt snappier than official 64-tick matchmaking, and third-party platforms built businesses on that gap.

CS2's answer was to go tick-rate-independent. Valve developed CS2 alongside new server architecture allowing for sub-tick gameplay that more accurately synchronizes with player input. Counter-Strike 2 - Wikipedia Steam markets it bluntly as tick-rate-independent gameplay. Counter-Strike 2 on Steam

In plain English: instead of sampling your click on the next 64 Hz or 128 Hz tick, the server timestamps the exact sub-tick moment you shot, moved, or threw. Peeks, counter-strafes, and jump-throws resolve in the order you did them, not the order the server happened to poll.

My contrarian read: sub-tick is an honesty patch. Valve did not win the tick-rate war by giving everyone 128-tick. It deleted the spec sheet. When netcode becomes invisible infrastructure, players stop shopping for servers by number and start blaming themselves again — which, for a competitive game, is retention gold. I find this braver than chasing 240 Hz marketing. It trades a bragging number for fewer excuses.

2. Dynamic smoke: from script to physics toy

Old smoke was a sphere with a timer. CS2 smoke grows to fill spaces, and can be altered in real time by gunshots or through the use of hand grenades. Counter-Strike 2 - Wikipedia Steam calls them game-changing dynamic smoke grenades. Counter-Strike 2 on Steam

Shoot through smoke and you carve a sightline. Throw an HE and you briefly clear it. Smoke seeps through doorways and fills a closed room instead of clipping through walls.

This re-prices utility. In CS:GO, a one-way smoke was memorization: 200 hours of lineups. In CS2, a one-way is a negotiation that the enemy can shoot open. My take: this favors improvisers over librarians. The team that practiced 50 set smokes loses edge to the AWPer who understands volume and timing. If you loved CS as a memory game, you will hate this. If you loved CS as poker, you just got more bluffs.

It also quietly nerfs the smoke wall and wait stall. Aggressive HE clears and bullet carving reward teams that interact instead of turtling. I think Valve wanted fewer 40-second stare-downs without admitting the pacing problem.

3. Maps in three buckets, not one remake

Valve did not rebuild everything equally. Many maps were upgraded to use Source 2 lighting and physically-based materials, grouped as Touchstone for layout-identical maps like Dust II, Upgrades for large-scale graphical upgrades like Nuke, and Overhauls for ground-up reconstructions like Inferno. Counter-Strike 2 - Wikipedia

That triage is engineering honesty. Touchstone preserves 20 years of grenade geometry. Overhaul admits Inferno's lighting never worked. For indie devs, the lesson is brutal: do not remaster uniformly. Spend polygons where readability was broken, freeze where muscle memory lives.

Add redesigned visual effects and audio, a revised loadout (five pistols, five mid-tier, five rifles), silent reloads that trade time for noise, and Premier mode with CS Ratings plus global and regional leaderboards — all listed on Steam as CS2 pillars alongside overhauled maps. Counter-Strike 2 on Steam None of these sell screenshots. All of them sell one more match.

What it cost: performance, macOS, and missing toys

Valve's reception summary is unusually candid: generally favorable gunplay and maps, but mixed player reception with criticism of performance, removal of several CS:GO features, and discontinuation of macOS support. Counter-Strike 2 - Wikipedia

Three costs deserve naming:

Higher floor, same ceiling. 85 GB and physically-based lighting pushed low-end PCs harder at launch, even with that modest i5 750 minimum. Competitive players who ran CS:GO at 300 fps on potatoes suddenly tuned settings again.
No macOS. CS2 ships Windows + Linux only. The CS:GO macOS client died with the upgrade. For a proprietary engine pitching Vulkan/Linux as future-proof, abandoning Apple is consistent — but it still cut off a slice of casual queue.
Features lagged. Arms Race, community modes, and Workshop tools did not all arrive day one. Replacing CS:GO outright on Steam — official matchmaking for Global Offensive was shut down upon release, though community servers remain and console versions stay playable — meant there was no fallback client. Counter-Strike: Global Offensive - Wikipedia Bold, but it turned every missing checkbox into a review bomb.

My take: Valve chose custodial pain over fragmentation. Keeping CS:GO alive alongside CS2 would have split inventory, ranks, and pro scene. Killing it forced complaints into one funnel where patches compound. I disagree with reviewers who call this arrogance. It is inventory economics: two queues halve the market that gives skins value.

Three ideas most coverage misses

These are my interpretations, not Valve statements. Take them as lenses, not facts.

1. The toaster strategy beats the ray-tracing strategy. Unreal showcases chase RTX. CS2's minimum — 1 GB DX11, i5 from 2009 — chases internet cafes and LAN houses. CS:GO drew an estimated 11 million players per month and remained one of Steam's most played games. Counter-Strike: Global Offensive - Wikipedia You do not keep that by requiring an RTX 4070. Source 2's Vulkan path lets Valve look modern on high-end (better lighting, PBR) without raising the entry ticket. Accessibility is anti-cheat: a bigger pool dilutes smurfs and shortens queue times more than any ban wave.

2. Silence as a feature. The new reload option — discard the mag now vs silent-but-slow — plus redesigned audio is not flavor. CS has always been an information game disguised as an aim game. Sub-tick makes when precise; silent reloads make whether you are heard a choice. I think future CS meta will be less about spray transfers and more about sound budgeting. Pros already manage footsteps; now they manage reload theaters.

3. Engine as archive. GoldSrc to Source to Source 2 maps 25 years: mod (1999) to retail (2000) to Condition Zero on GoldSrc to Source remake to CS:GO on Source in 2012 to CS2 on Source 2 in 2023. Counter-Strike - Wikipedia Each jump preserved maps, weapons, and recoil DNA. Valve never rebooted Counter-Strike; it re-housed it. That is why Source 2 has only a handful of games — Dota 2, Alyx, CS2 — while Unity has millions. Valve does not license an engine. It curates a museum you can still queue in.

What indie devs can steal (without Valve's budget)

Delete the spec war. If players argue about your tick rate, frame cap, or seed, you have a communication bug, not a tech bug. Sub-tick's lesson: timestamp inputs, explain it once in plain language (your click counts when you click), then never mention Hz again.
Make utility physical, not scripted. Even a 2D smoke that bullets thin out creates more clips than ten static walls. A little simulation beats a lot of content.
Remaster by bucket. Label maps Touchstone / Upgrade / Overhaul internally. Freeze competitive geometry, spend art where readability failed. Players forgive ugly if fair; they never forgive moved cover.
Carry inventory forward. If your sequel resets cosmetics, it is not a sequel — it is a competitor to your own game. Free upgrade plus item carryover kept CS2's economy alive on day one.
Ship low, scale with API, not assets. Target i5 plus 8 GB plus Vulkan, then layer PBR and lighting. Reviews reward 144 fps on old laptops more than ray-traced puddles.

Feel the ideas in a browser (no 85 GB download)

You cannot practice CS2 utility without CS2, but you can train the underlying skills that Source 2 rewards: crosshair placement, timing, and utility discipline.

For raw aim and pre-aim before you touch Premier, warm up in BuildNow GG's Aim Training and BoxFight. The game's own guide recommends static then moving targets, then tight-box peeks — the same loop CS coaches prescribe for counter-strafing. Short 5-minute routines transfer surprisingly well because both games punish wide swings.
For tactical patience — no respawns, plant/defuse pressure, trading kills — try Hazmob FPS: Online Shooter's Search and Destroy mode. Like CS, there are no respawns per round, so listening, saving utility for the push, and not re-peeking the same angle decide rounds. Its seven-mode structure makes it a good lab for learning when to chase kills vs play the objective.
For economy thinking — when to force, when to save, how damage scales — mess with Stick Merge. It is an arcade merge-shooter, not a tactical sim, but its core loop (merge to higher tiers, keep the strongest gun active, stack multipliers wisely) is a playful mirror of CS money management: bad buys snowball.

None of these replace CS2's recoil patterns or volumetric smoke. They let you rehearse decisions — peek discipline, utility timing, buy logic — in a tab while your main client updates.

FAQ

What engine does Counter-Strike 2 use?
Source 2, Valve's proprietary C++ engine. Source 2 - Wikipedia

What engine did CS:GO use?
Original Source. CS2 was produced as an update moving from Source to Source 2. Counter-Strike 2 - Wikipedia

When did CS2 release?
Announced March 22, 2023, released September 27, 2023 for Windows and Linux, replacing CS:GO on Steam. Counter-Strike 2 - Wikipedia

Is CS2 free?
Yes, free-to-play as a free upgrade to CS:GO, with an optional Prime Status upgrade for Prime matchmaking and drops. Counter-Strike 2 on Steam

What is sub-tick?
Valve's tick-rate-independent server architecture that timestamps actions at the exact moment of input rather than the next tick, for more accurate synchronization. Counter-Strike 2 - Wikipedia

What other games use Source 2?
Dota 2 (ported 2015), Artifact, Dota Underlords, Half-Life: Alyx (VR), Counter-Strike 2, Aperture Desk Job, and upcoming S&box and Deadlock. Source 2 - Wikipedia

My bottom line: Counter-Strike 2 is not impressive because Source 2 is powerful. It is impressive because Valve resisted using that power. Sub-tick deleted a decade of forum wars, dynamic smoke turned memory into improvisation, and carrying every skin forward turned an engine upgrade into a custody agreement. In a market addicted to new engines, Valve won by refusing to switch.

Why Roblox Bet on a Custom Engine and Luau: The Boring C++ Sandbox Behind 85 Million Daily Players

dev game — Wed, 21 Oct 2026 04:35:15 +0000

Why Roblox Bet on a Custom Engine and Luau: The Boring C++ Sandbox Behind 85 Million Daily Players

Short answer: Roblox runs on its own C++ engine plus Roblox Studio and a Lua fork called Luau — not Unity or Unreal — because it is not selling a game, it is running a mall for millions of untrusted kid-made games on phones, Chromebooks, consoles, and PCs at the same time.

That choice looks old-fashioned until you count the constraints: Roblox reported an average of 85.3 million daily active users as of February 2025, and says half of American children under 16 play each month. No licensed engine in 2006 offered free-to-play distribution, one-click multiplayer replication, built-in moderation, and a cash register that takes a cut of every transaction. So Roblox built the cash register itself.

What Roblox actually runs on

The company history matters: Roblox Corporation was founded in 2004 in Menlo Park after Baszucki's Knowledge Revolution made Interactive Physics and Working Model, started as DynaBlocks beta in 2004, renamed Roblox in 2005. It moved from third-party cloud to proprietary infrastructure after hiring from Dropbox in 2018 and buying mobile-network optimizer PacketZoom, then listed on NYSE as RBLX March 10, 2021 with $3.60 billion revenue in 2024.

That physics-toy lineage is not trivia. It is the engine spec.

Why not Unity or Unreal? Four boring reasons

1. Untrusted code at kindergarten scale.
Unity and Unreal assume a trusted studio. Roblox assumes 2 million+ developers, a majority minors as of 2020, uploading code that will run on a stranger's phone. Luau's standard library has no file-system or OS access by default, enforces memory and infinite-loop limits, and the GitHub docs push safeenv sandboxing to isolate scripts from each other. Luau.org frames this as embeddability: sandboxed by default so embedders can safely execute untrusted user code. No general engine gave them that in 2006.

2. One build must run on a school Chromebook and a PS5.
Luau rejected LuaJIT partly over limited console support and maintenance pain, and built its own VM instead. Benchmarks put Luau's interpreter near LuaJIT's interpreter mode, with LuaJIT's JIT about 1.6x faster in some tests — then Luau added native code generation in October 2023 for x64/ARM64 with 1.5x-2.5x gains on intensive code. That is a political compromise, not a benchmark chase: interpreter everywhere, native where allowed. Unreal's high-end-first scaling would have excluded the exact classroom devices that made Roblox grow.

3. The engine is the store.
If Roblox had shipped on Unity, it would pay a landlord while trying to be a landlord. With its own engine it owns replication, avatar catalog, Robux checkout, and the DevEx haircut. That is why templates ship inside Studio to let anyone publish and why more than 20 million games per year were reported in 2020. Friction is the product.

4. Physics-class DNA.
Baszucki did not come from shooters. He came from 2D physics simulation for education. A block that a 9-year-old can stack, drive through, and sell is more valuable than a photoreal shader that needs a technical artist. The chunky avatar persists because legibility beats fidelity when your creators are children.

What Luau actually changed from Lua 5.1

If you only remember Lua from Garry's Mod or World of Warcraft addons, Luau looks familiar but behaves like a stricter, faster cousin:

Gradual + structural typing with inference: add : number, : string or leave it dynamic. Strict mode aims to catch errors before runtime; non-strict minimizes false positives while migrating old code. Luau.org calls this safety: refinements and type functions powered by inference to rule out errors and drive autocomplete with minimal annotations.
New runtime, not a patch: C++ VM from scratch with constant folding, dead-code elimination, peephole optimization, and inline caching, compiling ~950,000 lines/sec on a Ryzen 5900X. GitHub stresses it is largely rewritten, preserves Lua 5.1 API for bindings, but separates compile (luau_compile) from load so you can ship a VM without a compiler.
Modern syntax without breaking 5.1: +=, string interpolation with backticks added February 2023, if ... then ... else expressions, continue, const + table.freeze.
Pause-free GC: Incremental collector with PID-controller heap pacing borrowed from Go, plus closure/upvalue/weak-table tuning to cut pauses. This matters for 30-player obbies where a stop-the-world hitch reads as cheating.
Tooling as first-class: Compiler, type checker, linter and language server shipped open-source, with luau REPL sandboxed and luau-analyze for CI.

The outside validation is telling: Remedy replaced proprietary scripting with Luau in Northlight for Alan Wake 2 and deleted ~80,000 lines, Digital Extremes moved Warframe for GC/memory wins, Giants adopted it for Farming Simulator 2025, and Linden Lab shipped SLua for Second Life in December 2025. GitHub now lists Alan Wake 2, Farming Simulator 2025, Second Life and Warframe as Luau users. Luau is quietly becoming the SQLite of game scripting — not the coolest language, but the only Lua fork that solved consoles + sandbox + types together.

My take: five ideas you will not find in the docs

These are interpretation, not Roblox statements. I label them as such because the task asks for distinct analysis, not recycled conclusions.

1. The type checker is moderation, not just DX.
Everyone praises Luau types for autocomplete. I think the real payoff is support-ticket deflection. When your creators are kids, a crash in front of 20 friends is a moderation event. Strict mode plus lint catches the nil-index before it becomes a 1-star review and a parent email. It is a chat filter for code.

2. Kindergarten-server theory.
Most engines optimize for a trusted team of 50. Roblox optimizes for 2 million strangers, many hostile or careless. Resource limits, no OS access, per-script globals — that is not paranoia, it is daycare architecture. Unity trusts you; Roblox assumes you will while true do end on someone else's iPad.

3. Lag as democracy.
Black Myth: Wukong bets on Nanite to show more. Roblox bets on running at all. Interpreter-first means a 2015 Android can still join Adopt Me, while native code lets a gaming PC run a 200-player shooter smoother. The ceiling is set by the poorest classroom device, not the richest GPU. That is the opposite of prestige-engine thinking, and it is why growth accelerated in the second half of the 2010s and again during COVID.

4. Anti-Unreal economics.
Royalty engines tax the developer. Roblox taxes the transaction. Owning the engine lets it give Studio away free, then take a cut of Robux, limiteds resale, and DevEx. You cannot do that cleanly on someone else's runtime without paying rent twice. The custom engine is a cash-register decision disguised as a tech decision.

5. Boring is a moat.
C++ core that never breaks old games, Lua 5.1 compatibility that never orphans tutorials, blocks that never need retopology — this is deliberate boredom. While every 2024 obby complains about graphics, no 2012 obby stopped working because the renderer changed. That backward compatibility is why parents trust it and why competitors with prettier demos cannot steal the library of 20M+ games per year.

If I were cloning Roblox today, I would not start with graphics. I would start with sandbox + payments + a linter that talks like a teacher.

What indie devs can steal (without building a country)

Sandbox before shaders. Ship safeenv-style isolation and per-script timeouts first. Kids and modders forgive ugly; they never forgive a crash that eats inventory.
Two typing modes. Offer non-strict for jams and strict for launch. You get fast onboarding and fewer midnight hotfixes.
Templates as growth. Roblox Studio's pre-made templates do more for retention than any ad. One obby template beats ten docs pages.
Device floor, not ceiling. Profile on the cheapest device your players actually own (often a Chromebook or old Android), then add native/JIT as a bonus, not a requirement.
Own the register if you host UGC. If players sell to each other, you need first-party commerce and DevEx-like payout rules from day one. Bolting a store onto a licensed engine later creates fee-on-fee pain.

Try the building intuition in your browser

You do not need Studio to feel why this loop works. These free browser games isolate one Roblox muscle at a time — and they run instantly with no install:

For block-break-collect-rebuild and Creative vs Survival pacing, play Mine Clone. Start small, gather wood/stone early, then plan a build — the same tolerate-failure loop that lets a 10-year-old publish an obby without fear. Its world-generation-every-time design shows why simple voxels beat high-poly when creators are beginners.
For decoration-as-ownership (the Adopt Me / Bloxburg house loop), try Craft World. With 300+ blocks and furniture layering, it teaches what Roblox learned: kids do not just build walls, they curate sofas, lanterns and themed rooms. That interior-design itch drives Robux avatar sales more than any gun stat.
For open-city roleplay plus missions-cash-shop (Brookhaven, Driving Empire in miniature), roam Sandbox City - Cars, Zombies, Ragdolls. Talk to Mission Guy, claim daily cash, buy a bat before a rifle, hijack any car with E. It compresses Roblox's mission → cash → shop → boost economy into a five-minute session.

If you enjoy the building-under-pressure side of Roblox BedWars, add BuildNow GG Freebuild + BoxFight drills: 90s, ramp rushes and box-peeks train the same place-edit-shoot reflex Studio PvP demands, but in a tab.

Limits and open questions

Performance still cliffs: interpreter-first saves compatibility but intensive physics still needs native code or clever throttling. Large servers shard aggressively for a reason.
Moderation does not scale with code: sandboxing stops os.execute, it does not stop social engineering. Scams around free Robux, beamers stealing session tokens, and once-per-account rollbacks remain documented problems.
Age-gating is tightening: ratings now split Everyone / 9+ / 13+ / 17+ with ID verification for 17+, and blocks in China, Turkey, Iraq and others show the country-scale cost of hosting kid UGC.
Luau's outside wins (Remedy, Warframe) prove portability, but they do not prove Roblox's networking and commerce stack is portable. The language is MIT; the mall is not.

Bottom line: Roblox did not pick a custom C++ engine and Luau because they are exotic. It picked them because no store-bought engine wanted to be a school, a sandbox, a phone OS, and a bank at once. Boring, sandboxed, and payable beats pretty — when your developers are eight and your players are 85 million.

Why Call of Duty Bet on the IW Engine: The Boring Quake-III Heir Behind 500 Million Copies

dev game — Wed, 21 Oct 2026 04:04:56 +0000

Why Call of Duty Bet on the IW Engine: The Boring Quake-III Heir Behind 500 Million Copies — And Why 60fps Beat Pretty Graphics

Short answer: Call of Duty runs on Infinity Ward's proprietary IW engine, a 20-year incremental fork that started from id Tech 3 (the Quake III Arena engine) for Call of Duty 2 onward. Infinity Ward stuck with it instead of switching to Unreal because COD sells feel — locked 60fps, low input lag, readable sightlines, and a yearly pipeline three studios can ship on — not screenshot realism. The current line is IW 8.0 for Modern Warfare (2019) and Warzone, and IW 9.0 for Modern Warfare II (2022) onward, unified across Infinity Ward, Treyarch, Sledgehammer and Raven.

That choice looks unfashionable next to Unreal Engine 5 or Frostbite. It is the reason COD still exists as a yearly product.

The question players actually Google

“What engine does Call of Duty use?” is one of the most repeated engine questions in shooters. Google suggests 10 variants — what engine does cod use, what engine does warzone use, what engine does call of duty mobile use, what graphics engine does call of duty use — and Bing suggests the same cluster plus iw8, iw engine, call of duty game engine. The franchise page confirms the scale behind the curiosity: over 500 million copies sold as of October 2023, 100 million monthly active players, $30 billion revenue by 2022, and Guinness verification as the best-selling FPS series ever.

The answer changed by era, which is why so many forum answers contradict each other:

2003 Call of Duty: id Tech 3 derivative. Infinity Ward was founded in May 2002 by 22 ex-2015 Medal of Honor: Allied Assault developers, funded with $1.5 million for 30% by Activision, then fully acquired in October 2003 the day after COD 1 shipped.
2005–2016 (COD 2 to Infinite Warfare): IW 2.0 through IW 7.0 — same lineage, renamed each generation.
2019–2020 (Modern Warfare reboot + Warzone): IW 8.0, described as “an entirely new engine” with photogrammetry, streaming and ray-tracing support, shipping October 25, 2019 and powering Warzone’s 150-player Verdansk.
2022–present (Modern Warfare II, Warzone 2.0, Modern Warfare III, Black Ops 6/7): IW 9.0, “powered by a new version of the IW engine,” released October 28, 2022 and the fastest COD to $1 billion in revenue. Activision calls this the unified engine — all studios on one base with shared tools, cross-play and one Call of Duty HQ launcher.

id Tech 3 itself lists the IW engine as its successor alongside id Tech 4, noting IW was used from Call of Duty 2 onward. That single sentence explains 20 years: COD never rewrote, it forked.

Why not Unreal? Three constraints most explainers miss

Most articles say “Activision wanted control.” That is true but vague. The specific constraints were:

1. 60fps is the product, not a setting

From Call of Duty 2’s Xbox 360 launch — 1.4 million units in year one with an 85% attach rate to new Xbox 360s — Infinity Ward locked 60fps on console while competitors targeted 30fps with better lighting. IW’s renderer, job system, streaming and audio mixer were all budgeted backward from that number: smaller maps with dense occlusion, baked lighting + selective dynamic lights, aggressive LOD, 60Hz server tick targets, and animation cut to preserve input latency.

Unreal could do 60fps, but Epic’s demo-driven roadmap in 2006–2014 prioritized visual features over COD’s exact latency budget. Owning the engine let Infinity Ward say no to a pretty feature that cost 3ms. Licensees cannot say no that easily.

2. Three studios, one calendar

Since 2005 Activision ran a rotation: Infinity Ward improves the engine for one game while Treyarch ships on the updated prior engine, then they swap. Sledgehammer and Raven later joined. That only works if the API is boring and stable.

A third-party upgrade — say Unreal 3 to Unreal 4 in 2014 — would have forced all four studios to retool at once and broken the yearly ship date that funds everything. The IW fork let them merge features incrementally: Treyarch’s Zombies systems, Sledgehammer’s traversal, Raven’s Warzone live-ops tooling, all back-ported into IW 8.0/9.0.

The 2022 unification was an admission this had partially failed — by 2020 each studio had drifted into its own IW branch, which is why Warzone integrations broke weapons and file sizes ballooned past 200GB. IW 9.0 was less a new engine than a forced re-merge.

3. Netcode and tools over renderer

id Tech 3’s snapshot + delta-compression + Huffman UDP netcode gave COD a head start on 12–18 player 60Hz matches, then 64-player Ground War and finally 150-player battle royale with cross-play and cross-progression. Warzone hit 6 million players in 24 hours and 100 million downloads by April 2021 on IW 8.0 — a scale test Unreal licensees rarely attempt with a modified shooter base.

Equally important: COD’s content factory — Gunsmith with up to 60 attachments per gun, seasonal weapons, operators, maps — needs artist tools that non-engineers can run. IW’s editor chain is ugly but purpose-built for that factory. General engines optimize for being general.

What IW gained — and what it cost

Gained:

Consistent gunfeel across 20 games. A Modern Warfare 2 (2009) player can pick up Black Ops 6 and aim correctly in 30 seconds.
Scalability from 6v6 to 150 players without changing engines. Warzone’s Buy Stations, Loadout Drops, Gulag 1v1 respawn and Plunder economy were added as modes, not rewrites.
Live-ops speed: seasons, weapon tuning, anti-cheat updates ship without waiting on Epic or DICE.
Photoreal-ish reboot without losing speed: Modern Warfare (2019) reworked multiplayer for tactical play, added 100+ player Ground War, Gunfight 2v2, and sold over 30 million units by September 2020 while keeping 60fps on base PS4/Xbox One.

Cost:

Visual ceiling. Frostbite’s destruction (Battlefield) and UE5’s Nanite/Lumen still embarrass IW in tech demos. COD maps are corridors with skyboxes, not fully destructible cities.
File size and tech debt. Three merged branches + 4K assets + Warzone = infamous 200GB+ installs.
Single-player compromise. Modern Warfare II’s campaign is praised but its water physics, swimming and vehicle hijack systems feel bolted on compared to a purpose-built open-world engine.
Talent moat that cuts both ways. Twenty years of internal knowledge makes hiring fast for COD, slow for anyone else — new engineers must learn a Quake-III-descended codebase no school teaches.

My take: the engine is a calendar, boredom is the moat, and “dated” is strategy

Here is where I diverge from the usual “COD engine old, Unreal better” consensus. I think three contrarian points explain COD better than any renderer comparison:

1. IW is not a renderer, it is a release calendar compiled to C++.

Unreal asks “what could you render?” IW asks “can three studios ship every October at 60fps without crunching the tools team to death?” That reframes every “why not switch” debate. Switching engines would not just change graphics — it would cancel a $30-billion yearly cadence for 2–3 years. No board approves that to get nicer reflections. Frostbite’s mandate at BioWare (Dragon Age, Mass Effect) proved the opposite: forcing an FPS engine onto RPG teams for corporate synergy broke games. Activision avoided a mandate by letting each COD studio keep its IW branch — until drift forced the IW 9.0 re-merge. Boring incrementalism beat grand unification for 15 years.

2. 60fps is not a tech spec, it is a brand promise that filters the audience.

Most critics compare still screenshots. Players buy motion. COD’s low time-to-kill (150–300ms) only feels fair at 60fps+ with low latency. Drop to cinematic 30fps and every death feels like lag, no matter how good Lumen looks. My read: Infinity Ward correctly treated 3ms of input latency as more valuable than any lighting feature. That is why “COD looks dated” never hurt sales — readability is the feature. Enemies must pop against the background at 100 meters while sprinting. Photoreal clutter would make the game worse.

3. The fork-don’t-rewrite doctrine is the real lesson, and Warzone proved its limit.

Twenty years of forking Quake III sounds insane until you list the alternative graveyards: id Tech 5/6 pivots, Frostbite RPG rewrites, Slipspace for Halo Infinite. IW survived by never rewriting. But Warzone exposed the bill: when Verdansk, Rebirth Island, Caldera and Fortune’s Keep had to share one client across Modern Warfare, Black Ops Cold War and Vanguard content, loadout and weapon-balance parity collapsed. Unified IW 9.0 + Call of Duty HQ was not innovation — it was debt collection. Indie devs should copy the first 15 years (fork, don’t rewrite) and heed the last 5 (schedule the re-merge before three branches become three engines).

If I were advising a small FPS team, I would steal COD’s order of operations: lock tick rate + target fps first, lock map size + player count second, choose renderer third. Most indies do it backward.

What indie devs can steal this week

Budget backward from feel. Pick 60fps or 120fps on your weakest target hardware, then cut features to hold it. Do not add Lumen-style features and hope optimization saves you.
Build the Gunsmith, not the gun. COD’s 60-attachment tuning system looks like content, but it is retention: players theory-craft instead of churning. A tiny attachment + stat-bar system beats 10 more guns.
Design modes as content, not code. Gulag, Plunder ($1M cash race), Resurgence (respawn while squad lives), DMZ extraction — all shipped as modes on IW 8.0/9.0. Prototype modes first; engine features second.
Unify late, but unify. If you fork for DLC or co-dev, schedule a merge milestone. COD waited too long and paid in gigabytes.

Train these COD fundamentals free in your browser

You cannot practice Warzone’s 150-player rotations without installing 200GB, but you can drill the underlying skills that IW rewards — positioning, loadout switching, and objective discipline — in lightweight browser shooters.

If you die because you take fair 1v1s in the open, run BuildNow GG on EasyHub. Its five modes — Freebuild, V-Arena, BoxFight, Zone Wars and Aim Training map directly to COD drills: Freebuild for 90s and high-ground takes (think rotation to circle), BoxFight for edit-peek shotgun timing (think Gulag 1v1s), Zone Wars for final-circle positioning with a shrinking zone. The page’s sensitivity 3–6, FOV 90–100, and shotgun-vs-rifle-by-range advice is the same logic as COD Gunsmith range discipline.

If you lose because you chase kills and ignore objectives, run Hazmob FPS: Online Shooter. Its seven documented modes — Team Deathmatch, Free for All, Capture the Flag, Domination, Search and Destroy, Elimination and Gun Race with 55+ customizable weapons force the Domination/Capture-the-Flag lesson COD players learn late: standing on the point with teammates wins more than distant kills. Start with Team Deathmatch to learn one map’s lanes, then Domination for rotations, then Search and Destroy for no-respawn patience — the same progression from 6v6 to Search to Warzone.

Neither browser game replicates IW’s netcode or recoil, but both train the decisions IW punishes: open-field peeks, wrong gun at wrong range, and solo pushes without cover. Fifteen minutes of BoxFight peeks + one Domination match is a better warm-up than another YouTube loadout video.

FAQ

Does Call of Duty use Unreal Engine?
No. Mainline COD has never shipped on Unreal. It shipped on id Tech 3 derivatives, then IW 2.0–9.0. Call of Duty Mobile uses Unity (TiMi), Warzone Mobile uses a mobile-adapted IW branch — which is why mobile feel differs.

What is the current COD engine?
IW 9.0, introduced with Modern Warfare II (2022) and shared across Warzone 2.0, Modern Warfare III (2023) and Black Ops 6 (2024) via the unified engine + Call of Duty HQ launcher. IW 8.0 powered Modern Warfare (2019) and Warzone 1.0.

Why does COD still hit 60fps on old consoles?
Because the engine budgets everything else around it: map size, player count, lighting, and file size all yield to frame time. That is the Quake-III inheritance — snapshot netcode, delta encoding and lean OpenGL roots optimized for speed over spectacle.

Should indies copy IW?
Copy the doctrine, not the code. You cannot replicate 20 years of tools, but you can replicate the discipline: lock fps, fork boring tech, build artist tools early, and prototype modes before rewriting renderers.

Engine history moves fast: Modern Warfare 4 is scheduled for October 23, 2026 and Black Ops 7 launched November 14, 2025 per the series timeline — expect IW 9.x iterations, not a rewrite, unless Activision announces otherwise.