- For Talos Principle 2, Croteam switched from their own engine to UE5. The description was "It would be like attempting to sprint and catch up with a train that is already far down the track and accelerating even faster.". From a user's perspective, observe the graphics of Serious Sam Siberian Mayhem and Talos Principle 2: Same company, released at a similar time. Talos Principle (UE5) looks dramatically better than SS. (In-house engine)
- Jon Blow rolls his own engines (and lang), and releases games very slowly
- Expedition 33 recently released as a phenomenal game that leaned heavily on features in UE5 (Face/body, graphics, map/terrain gen etc.) They focused on the game itself, and let the engine do the heavy lifting... to a superb result.
From my own anecdotes in graphics programming: Producing a simple engine is easy. Producing something that's photorealistic etc is massively more difficult. Let along all the other things an engine provides for you. Modern games have so many complexities the engine abstracts out; we don't need to roll a new engine for each game or studio, each trying to have optimized netcode, human characters, photorealistic lighting, GUI map editors + terrain gen etc.
I use my own graphics engine for my scientific programs, but it has much simpler requirements than a game engine.
Very different world from 20-30 years ago where the initial "game design" was happening in paint programs or even graph paper notebooks.
Otherwise it can be a dangerous fool's errand on which many projects go to die. My younger naive self can attest to this, he loved trying to build his own overly-ambitious engines. But he never finished any games.
Another thought if you do roll your own - keep it simple stupid. When your brain tells you that some amazing nested scene graph with integrated occlusion culling would be the coolest thing in the world, but you lack evidence that you'll actually need all that functionality, tell your brain that it's being stupid and just implement some kind of basic flat scene structure. You can always retrofit it later.
Also - study the code of the likes of Carmack. Consider that he produced the likes of the quake engines in only a couple of years. Reflect long and hard on the raw simplicity of a lot of that code.
Do not worship complexity.
These are the words of someone who has walked both roads!
The same goes for software libraries in general, I think. Just make your program. Don't make an overly general library for something you won't need anyway. If the code proves useful for reuse, just factor it out after the fact and generalize as needed.
EDIT: Typos, wording
Things like the famous fast inverse square root are short, but I would hesitate to describe it as simple.
Ironically one of the things that the Quake engine relies on is clever culling. Like Doom, the level is stored in a pre-computed binary space partition tree so that the engine can uniquely determine from what volume you're in what the set of possibly visible quads is (if my memory is correct, oddly the engine uses quads rather than triangles) AND how to draw them in reverse order using painter's algorithm, because the software renderer doesn't have a z-buffer.
https://www.fabiensanglard.net/quakeSource/quakeSourceRendit...
The BSP partitioning used to take several minutes to run back in the day.
Anyway, the point I was trying to make was that Carmack used a few, clever, high-impact techniques to achieve effects, which were also "imperfect but good enough".
If you're not Carmack, don't over-optimize until you've run a profiler.
Not the best example. That snippet was in use at SGI for years and actually written by Gary Tarolli. Quake's optimization was mostly done by Michael Abrash.
The original id engines were also famously inflexible. They fit the mold of "developing an engine, not a game" to a T. What you saw them do was all they could do. Look at how much Half-Life needed to add to be viable. idtech3 also only broke out of its niche because Ritual and Infinity Ward heavily modified it and passed it around. There's a good reason the engine-based ecosystem is so prominent now.
I'm also working off a near 30-year-old memory but I recall quads not being unusual around this time. I remember a preview of Tomb Raider 3 in Official Playstation Magazine making a big deal out of the updated engine using triangles instead of quads to draw things. This was around 1998, so a couple of years after Quake came out.
Then you start it hit the more tedious stuff. loading animated characters, blending animations on selective subtrees of a character hierarchy. Making a level editor. Adding quality of life feature to it like undo. Etc…
I’m not saying you shouldn’t do this. It’s fun to do. just don’t delude yourself that that’s making progress on your game. It’s instead making progress on a game engine. That’s a different thing.
I've shipped 18 games, 4 of them AAA. I wrote the engines for most fo them. I wouldn't do it again.
All that said, some nuance. If the game you are making is simple for some defintion of simple, Celeste, Dead Cells, Geometry Wars. Then making your own engine isn't much work and there maybe some benefits.
On the other hand, see all the tiles made with engines. Silksong is Unity. A Short Hike is Unity. Blue Prince is Unity. Valheim is Unity. Peak is Unity. Dredge is Unity. You don't need to make your own engine to make an indie game.
"I thought if I made a really good engine, making a game would be the easy part!" I had similar thoughts when I was younger. Surely if I just upgrade my tools, the hard part will become the easy part!
Jonathan Blow says making engines is easy, because enginedev only takes a relatively small part of development — the game itself takes way more time and energy.
So his argument is, in the grand scheme of things, the engine is not that much work. (Since you're gonna spend ten years working on the game anyway, of course ;)
Programming an engine requires dedication, but pretty much every other area in gamedev require similar dedication to get to an acceptable result.
Solo? Or part of a team?
For certain personality types I think making an engine can make it very easy to get distracted and wind up in the weeds of something you don't actually need, overoptimizing, fence-painting etc. Using an engine can help with self-discipline and focus on the end rather than the means, although then you need to make sure you don't just wind up with a ton of mostly finished tutorial projects and no game.
they used NeXT workstations to develop it, the programming tools on PCs were too weak for such a project
today it might look simple, but it's easy to say that when you open it in VS Code and have Intellisense, autocomplete, go to definition, ultra fast compilers, tons of RAM, and google for everything
Also says something about the accumulation of complexity. At that time Carmack (and his team) were able to create a state of the art engine in a few years. Now consider the task today, if you were to create a state of the art engine today it'd take tremendously more work.
And yet often the actual gameplay code itself may only be 2x to 3x more complicated then the days of old.
I think of counterstrike for instance - it's still just guys shooting guns in a constrained arena.
So having AI build the slop instead of a human seems to make sense. I really wonder how AI is changing the gaming industry.
The articles author strangely left AI out of what he wrote. While I know a lot of HN readers are traditional and love the old way of doing things I don’t know how much longer that way will last.
For plenty of reasons there are lots of programmers who still need to know how to code in assembly, read assembly, etc.
Creating my own engine was both a personal and strategic decision for me. I was really worried about running into performance issues with generalist engines, and I did not want the friction of working with someone else's mental model. Pretty sure that friction would have caused so much burnout for me. There's also the long payoff of operating in an environment that you understand top to bottom.
I ignored all the advice about making smaller games first, creating an engine first, etc. Metropolis 1998 is my first game and so far it's working out just fine. But your mileage will vary.. I started development with 10+ years of software experience and fond memories of Rollercoaster Tycoon and SimCity 2000/4.
I only add what I need. There's no level/scene editors (outside of the game being one itself :P ). No scene graphs. Shaders are coded by hand. Right now the entire game is about 45MB.
[1] https://store.steampowered.com/app/2287430/Metropolis_1998/
(Also thanks this is exactly the kind of game I'm into.)
- - Total days - 1631
Engine: 610
Pathing / Roads: 287
Features: 477
Marketing: 45
Burnout: 130
Other: 82
You're welcome!
I was really enjoying reading this piece until I read the above, then I realized I am reading for a big developer, the maker of, Celeste [1]. I am definitely adding this to my list of favorite articles about making games.
Also, you may want to check a previous discussion from nine months ago (573 points, 246 comments ): https://news.ycombinator.com/item?id=44038209
_____________
Also, pretty sure it was a small indie team rather than a “big developer”
--
0: https://cityofnone.com/And any new stuff regarding Celeste or from their devs will forever be relevant to me! Highly recommend to any who haven't played it.
If I remember correctly it was a team of 2.
0: https://maddymakesgames.com/
Fans of Celeste will almost certainly enjoy the local multiplayer game Towerfall by the same developers.
I liked when games all felt very distinctly different and I feel like part of that was that they all varied on ‘engine’
But I suspect that when you have multiple years to build Tetris, you can spend a lot of time crafting your own style.
> AI tools will allow the best to reach even greater heights, while enabling smaller teams to accomplish more, and bring in some completely new creator demographics.
No need to brag about that.
Like what? If you can already program your game and create art for it, what is it going to be doing?
People are so obsessed with using AI stuff for the sake of it, it’s nuts
1. advanced autocomplete -- if you have or paste the structure of a JSON or other format, or a class fields, it is good at autocompleting things like serialization, case statements, or other repetitive/boilerplate code;
2. questions -- it can often be difficult to find an answer on Google/etc. (esp. if you don't know exactly what you are looking for, or if Google decides to ignore a key term such as the programming language), but can be better via an AI.
Like all tools, you need to read, check, and verify its output.
JetBrains also has local line-based LLM models for various languages.
With the LLM-based autocomplete it a) generally autocompletes more code at once, and b) will often pick up on patterns in the existing code. E.g. if you have a similar method, list of print/string buffer write statements, or other repetitive code in the file it will often use that as a model for the generated code.
If I have a JSON structure, I can paste that into the file as a comment, e.g.:
# {"foo": 1, "bar": "test", "baz"}
@dataclass
class FooBar:
foo:
@dataclass
class FooBar:
foo: int
bar: str
baz: int
def __str__(self):
return f'Foo(foo={self.foo}, bar={self.bar}, baz={self.baz})'
class Foo(Base):
__tablename__ = 'foos'
bar_id: Mapped[int | None] = mapped_column(ForeignKey('bars.id'), default=None)
bar: Mapped[Optional['Bar']] = relationship()
I guess they might finally get me to use those things since they take the “configuring” and “remembering shortcuts” part out, but so much of this doesn’t look new at all. Super old, actually.
With ai it add several lines of code at once as soon as it thinks it recognizes a common pattern.
It’s not perfect and it can get in the way but it’s amazing when it guesses right and spits out the next 3-4 lines I would have typed
Does the calculator give you a slightly different answer each time, even with the same inputs?
But at the moment it's also helping me solve more complex issues with building applications - it's JS, so you can imagine how complex it can be.
I yearn for a simpler workflow to be honest, I don't want to rely on SO or LLMs to solve build issues. I want to work in Go but there's only a handful of companies using it in my country, plus my CV basically says I mainly did front-end in the past ~15 years.
This is a GREAT observation. Thank you!
• There is overhead in learning how a specific game engine works.
• Often, due to a game engine API, it seems to herd you into writing the same game everyone else is writing with that engine.
I wanted just enough "game engine" to abstract away the pixel-buffer, windowing, user-events on the various target platforms and then do no more.
"I have been using SDL3 as it does everything I need as a cross-platform abstraction over the system - from windowing, to game controllers, to rendering."
And that is exactly where I landed as well. SDL3 [1] absolutely matched what I wanted. Then again, I enjoy writing sprite-based games. If you want to write a 1st-person shooter though I'm sure you will still want to go with one of the giant game engines.
(Actually it was SDL2 since it was two years ago I was exploring it: https://store.steampowered.com/app/2318420/Glypha_Vintage/)
[1] https://www.youtube.com/playlist?list=PLlrATfBNZ98drHSOb-h2e...
Godot, Unreal, CryEngine, and even Unity... all solve edge-cases most don't even know they will encounter. Trying something custom usually means teams simply run out of resources before a game ships, and is unlikely stable on most platforms/ports. =3
But even more complex custom/inhouse engines are usually not written from scratch, those are often mostly glued together from specialized middleware libraries which solve the tricky problems (e.g. physics engines like Jolt, vegetation rendering like SpeedTree, audio engines and authoring tools like FMOD or WWise, LOD solutions like Simplygon, etc etc...)
Most game engines are broken by default. Modern customers just aren't very discerning ("It's for the pigs. Pigs eat slop."). You can feel holes and rough edges in the vast majority of new releases, including AAA titles.
Unreal is the worst for this and Unreal-based games almost always have two things in common: a very particular, soft, sticky and unresponsive look & feel (often alleviated but never fully corrected by turning off some combination of motion blur, AA and VSync), as well as a UI that mishandles mouse pointers.
Unity devs seem to rely on a (more diverse but still quite) small pool of subsystems and renderers; possibly some mix of baseline and Asset Store components. This gives each Unity game a specific subset of flaws from a wider common pool. That is, you can tell that game A uses the same movement subsystem as games B and C (but not D), that game B uses the same UI subsystem as games C and D (but not A), and that game D uses the same rendering subsystem as games A and B (but not C).
Forcing devs to use a mid-grade GPU also tends to reduce chasing performance issues later. For example, high frame-generation artifacts users often perceive as "floats" or "wobbly". =3
Besides not all games need performance, I have been working on a clone of the original elite game using SDL. It gets 6000 fps easily.
[0]: https://ookigame.com
I could never jell with C++ until I had Cursor hold my hand (especially around the build system), and now I feel like I am developing games with a razor sharp knife that I could never before access. The patterns associated working directly with memory suddenly clicked and now it’s the only I want to build games.
I still don't use it (AI) for the game programming as it sucks the joy out of it for me. Especially when AI usage is currently being pushed hard at work.
But I haven't reached to the more tedious parts, like doing skeleton animation on GPU and testing cross platform (SDL should be naturally cross platform, but I never tested it...), etc. The most tedious part, imo, is to build your own 3D scene editor.
At very least I can say SDL has reached a passable state for 3D. It doesn't support some modern features like bindless though.
And one doesn't need to stich with C++ if they don't want to. SDL is pure C and if your favorite language can call foreign function it's not that hard to make it work with SDL.
I think that if I were making something in 3D or I were more serious I'd use an engine, but I've found that I get more satisfaction from building tools than from learning how to use tools that other people have built.
compiled a personal webpage to play with mobile controls and a javascript engine to play the pico-8 games and i love the celeste port on that!
The pro license at ~$2k/year per seat is all you need unless you are making a shitload of money. In which case, you are going to pay ~5k/year per seat.
I last looked into the matter when considering RFP's for government contracts for VR software. Didn't feel like haggling with Unity's sales reps, especially since the government hasn't been the greatest client of late.
All of this is before you get to the Asset Store, which largely seems to assume that gamedevs are the customers. I'd rather not re-read the license agreement for every asset I've bought, but I know for certain that a number of them are explicitly games-only.
I'm also not sure if it's still in the installer, but it used to ask you what you would be using unity for, and I don't remember most of the options, but one of them was "military simulations" or something like that, so they are aware of the possibility
So I narrowed my game concept to be more stylized, no photorealism, no human characters, etc, in order to make something both unique, and deliverable.
Its inspiring to see others doing the same thing for similar reasons. Maybe I'm not completely crazy after all :)
- Desktop: Windows, MacOS, Linux
- Mobile: Android, iOS, iPadOS
- Console: Playstation 4, Playstation 5, Xbox One, Nintendo Switch
It used to be XNA but then Microsoft discontinued and the community created the API compatible MonoGame.
Notable games: Terraria (when it was XNA), Stardew Valley, Celeste, Terraria and Fez.
React Native uses Skia under the hood as far as I recall.
I'd really love to go all-in with C# and SDL3 to make an engine-less cross-platform game but I still miss a good way to make complex game UIs without doing everything from scratch. Does anyone have a good suggestion for making non-trivial game UIs without using a full game engine? So far, I only found https://github.com/vchelaru/Gum and https://github.com/mikke89/RmlUi but I think there's not really something super mature and feature packed, or is there? I'm aware of https://github.com/ocornut/imgui, as the article also mentioned, but that's more for debug UIs, right?
Unity indeed is still using Mono instead of CoreCLR and is kinda stuck in that sense. But to be fair, they are trying to migrate to CoreCLR which will let them profit from all the crazy optimizations that Microsoft has poured into the runtime and ecosystem.
Godot is kind of a hate love for me when it comes to C#. Godot gives me the most hope that there can be a free, community-driven but powerful game engine and it having C# support built-in seems great at first glance until you realize that GDScript, which is veeeeery dynamic language, pretty much nullifies a lot of the advantages you'd get from using C# because you find yourself doing weird type system stuff that GDScript imposes on all the other languages. The best you can do is doing as much as possible in C# and use Godot as kind of a input and rendering abstraction layer. But then you're missing out on a lot of the functionality that Godot offers which should raise the question why use a game engine in the first place. It's difficult, at least for me. Others might have figured it out much better.
But I'm still not too enthusiastic about having GC in C# which is why ideally I'd like to start making a small 2D game just with SDL3 and C++ but how could I get this nice hot reload workflow there? I don't have the money to pay for expensive proprietary tools like https://liveplusplus.tech so what can I do? I guess there's the "game as dynamic library" trick from Handmade Hero (see https://www.youtube.com/shorts/seWAIURXxH0) so maybe that would work good enough? Maybe https://github.com/fungos/cr would do most of what's needed here?
Also, how does one even do modern C++ these days? Is it possible to have big C++ code bases that still compile fast these days? Is CMake 4 good™ now? Are modules really there yet? I rely on clangd as LSP for working with C++ but I read that clangd still fundamentally struggles with C++ modules https://chuanqixu9.github.io/c++/2025/12/03/Clangd-support-f... and it's so bad that there has even been some effort going into making a new C++ LSP https://github.com/clice-io/clice
I always thought CMake was good enough. I use FetchContent for all external dependencies and git submodules + add_directory for all internal dependencies. It took me a while to figure out that this was the simplest thing that works reliably though. I have used vcpkg and conan extensively and have completely given up on them.
I don't use C++ modules, because afaik they are still mostly broken. Without modules clangd works wonderfully and has been working wonderfully for a few years. I really, really like it. I think it can be done! I am missing reflection though, but if I would really want to use it, I'd probably just use clang -ast-dump instead of the new reflection functionality.
> I quite like modern C++, so I really tried to make it work for gamedev, but I think you can't really do it.
What exactly do you mean? What parts of modern C++ did not work for you?
> You can use all the modern C++ features that are not in the standard library though!
> I just decided to not use any of the C++ standard library (I just use a few C headers)
So what do you do with C++ that C alone couldn't do?
What I do like and use is overloading, references, templates, concepts, lambdas, enum classes, user defined literals, constexpr, operator overloading for math (!), little syntax stuff like structured binding declarations, "auto" and range based for. I also made my own little fmt (like https://riki.house/fmt). C++ is a much nicer language than C imo, even without the standard library.
It's just that I quite dislike using such a scripting language. It's personal preference, for sure, but here's a bit more context https://news.ycombinator.com/item?id=47220602
I think even `dotnet watch` at some point nopes out when you change too much. I think they call it "rude edit" and ask you to completely restart the program in that case. So I don't expect every possible C++ edit to be manageable by hot reload. But changing a few if conditions or constants should be fine or not?
I'm more and more questioning scripting languages in games. What are the main reasons to use something like Lua? I think it's having not to rebuild the engine, no compile times, changing stuff while the game is running and being more accessible to non-programmers. But I think it's rather infuriating, all those points could be less relevant if the tooling for "real" programming languages was better. And with coding agents becoming more wide-spread I guess accessibility to non-programmers also becomes less of a point. I guess it's just my personal dislike for scripting languages in games, but really, it would be so much nicer imo if there was only need for one language that does it all. But seems like a difficult thing to achieve.
Builds: https://github.com/hez2010/Satori/releases
how to use? do self-contained publish (but not single file), replace 3 files in the folder with the one from Satori release you can check if it's in use with GC.GetConfigurationVariables().ContainsKey("SatoriGC")
It is a far, far superior experience to touching anything C++.
its kinda insane how much i can get done with codex
half way through i realize i need a way to build levels and codex suggests you need a editor, here's the PRD following the safety harness and framework in your AGENTS.md and I come back to get coffee and its done....
also have a appreciation for how much game engines do for you, every step of the way I want to add a simple feature and realize we are not using a game engine so those have to built as well
If journey is more important to you than the destination then developing games without an engine can be a great adventure.
But if you bank on shipping your product within budget and scope then you'd better pick up one. Any one. And stick with it.
It’s easy for me. I just know C and raylib’s API is small. I got cross platform compilation going in an afternoon.
I’ve worked through some things with Godot. There’s just so much to learn that it’d take me longer to learn Godot than to get running with C.
Feel much the same as the author.
(246 comments)
I think the issue was when I used an engine the scope was too large and I never completed the work so I never released the game (or I released it for free because I felt it was incomplete and wasn't worth charging for)
It's great to work in a constrained environment
I saw this documentary on how celeste was made [1], which completely inspired me and got me into indie game dev community. Unfortunately I haven't made any games as of now that I would proudly showcase but the seed that your effort put is still there and one day I will get back to making games. Thanks a lot for making celeste I absolutely love it! ---- 1. https://youtu.be/MSKOQr_YS-U?si=AGzl5ILzxkoIB-j9
It's kinda sad SFML never get quoted, It was my framework ( after ALLEGRO ) where i learned c++ and I think it dosen't get much love nowdays even if it is very light and strong
- A former VS Code enjoyer
But yeah, it is like learning a new language. But that's not a bad thing! I found messing around / following some tutorials for e.g. pico-8 to be both liberating (two characters are enough for a variable name) and educative (using functions like min/max to the fullest)
I have wanted to make games my whole life but I got into web to make a living. Now it's been decades of game ideas and no implementations, just frustration. Something about unity, unreal, godot just never clicks with me
Maybe I'll try making my own engine too. Best of luck to you, maybe the both of us can make our dreams come true with a slightly different approaches
While most of the stuff is one off assets that do not fit together there are also some nice sets by some creators such as Kenney or Emcee Flesher
Also the Liberated Pixel Cup (LPC) stuff is pretty nice.
Mind you I mostly just look for 2D assets.
https://opengameart.org/users/kenney
I honestly don't see anything wrong with using the engine for its UI and "some rendering" kind of sweeps a lot of the complicated 3d light handling under the rug. I think the biggest mistake large engines have made is baking in features as first class citizens instead of those features being part of a standard plugin you could have written yourself from scratch once you reach that stage.
I've contemplated building my own editor UI, but after four weeks I realized that I'm just rebuilding the same UI structure you see in FreeCAD, Blender, Isaac Sim, Godot, etc. There's always a 3D viewport, there's a scene tree and there is an inspector panel for looking at the properties. So why not just use the Godot editor as a UI and provide my own custom nodes? By the time I've outgrown the training wheels, I've spent months working on the actually differentiating features.
Godot is more or less built that way. The entire node system is an abstraction over the various “servers” and you can even completely forgo it if you want, providing your own MainLoop implementation (instead of the included SceneTree implementation of MainLoop) and then just use the servers directly.
A lot of the editor features are also implemented as plugins. It’s been very easy turning the Godot editor into a custom editor for my game by writing some simple plugins
In theory, yes. In practice 99.9% of the games developers want to make are feasible with an off the shelf engine.
Seems like if you're doing this for a hobby or solo/small team then maybe it's reasonable.
For most people where they want to be a game dev but they probably will just work in industry, it seems like learning the major engines to competency cannot be ignored.
Heck, I've seen someone build a visual novel-type game with WinForms. That was actually a sensible choice for the game's presentation and interaction needs.
Of course if you want to become a game dev at a studio then you should be competent with whatever the studio uses (or something comparable so you can pivot to their stack). If you only want to make your hobby project and maybe publish it later it doesn't matter if your engine is Unreal, MonoGame, RPG Maker 2000, or vanilla JS/DOM.
I would say that one of the "Miscellaneous Thoughts" at the end of your article answers your question pretty well:
> I need only the best fancy tech to pull off my game idea
Then use Unreal! There's nothing wrong with that, but my projects don't require those kinds of features (and I would argue most of the things I do need can usually be learned fairly quickly).
Rust (the top 10 most downloaded game ever on Steam) is built with Unity. However they ended up to write their own netcode anyway. Of course Unity isn't known for the best netcode, but how much an engine helps is often overstated. Genshin even bought Unity's source code to customize it.
What's wrong with Netcode for GameObjects, and what are the odds I'll regret going with it?
Im sure NfG is fine
This is the mantra of the past decade of game dev.
that's probably what you mean by highest level of complexity
but even just a regular turn-based game there's a lot of questions you need to answer and decisions to make regarding latency, packet loss, and disconnects that all have to be solved in a coherent and consistent way