Unity compiles down to 3MB wasm. Though they don't need to ship the dotnet runtime, they use IL2CPP, a proprietary thing that translates dotnet code into C++. Then they run Emscripten on it.
Also I think Godot's WASM outputs used to be 2-3x smaller in Godot 3 (though the C# one was bigger).
You can ship a JS game in kilobytes, although atop 30 million lines of Chromium that's cheating of course :) still good fun.
Docker can be small too. In this example I was able to compile a full server (rust binary) and package it in a docker (scratch image base) and the total was < 5MB.
Seems like the anti-clickbait title editing removed the “hundreds of”? This confused me because I thought “oh, they’ve stripped down a Docker image to only 1 megabyte, and then a full game engine adds 34MB more” (I missed the WASM on the first read)
Meanwhile: Error - The following features required to run Godot projects on the Web are missing: WebGL2 - Check web browser configuration and hardware support
This is on Chrome 148.0.7778.96 (Official Build) (64-bit) on Fedora 44, 14900K, 4090 RTX, 128GB RAM
35MB WASM which relies on browser, which relies on drivers and OS, AND it doesn't work as advertised. I get the point, but point was mistaken.
It is a common argument when sizecoding is discussed. "You say it is 64k, but I need a multi-gigabyte OS to run it...". 35MB can't hardly be called sizecoding, but that's the same idea.
It is true, but besides the point because we have no choice. It is not the 1980s anymore and we can't access the hardware directly on personal computers, vendors don't let us. Not that we would want to, for portability reasons. So we depend on abstraction layers, which are not particularly lightweight, and here, it is a working browser supporting WebGL2 and WASM. You simply can't have a game engine running on modern hardware without a software platform. For the sake of argument, we could imagine a lightweight (tens of MB or less) WASM bare metal platform with simple pixel pushing hardware.
WebGL2 has been supported in Chrome since 2017.[0] The last holdout, as is tradition, was Safari which gained support in 2021. MDN considers it "Baseline - Widely available."
no offense taken. Of course it's my setup, even though it's vanilla setup on a new driver where driver and gfx do work. That's the thing illustrated here - it's always someone's setup somewhere outside of general case tested and "works on my machine". Then you get into all these setups and see variations and one day you decide for common libs to be shipped with your game (remember those MSVC installs?), then still same issues and then you decide to ship common runtime (like steam has), then you decide to ship hardware.. on more constrained (SLAs and whatnots) side it was the same thing and that's how we ended up with docker images.
Shameless plug but if you like small file size overheads, our browser based game engine Construct[1] exports an empty web project with about 300 KB overhead for a fully-featured engine. We achieve this by going all-in on the web platform so we don't have to ship a heavyweight runtime with it, and using a modular approach where only the components you use get exported.
It's not an error, it's just saying the project exceeds the free edition limits, which means you can look around and preview it but not make edits to it.
So the engine is 35MB, but what does it run in? By itself it's a binary blob that doesn't do anything.
Chrome seems to be ~404MB installed on here; that is conspicuously missing from the comparisons here to Docker containers which do account for more or less the complete runtime.
That is a strange title / comparison. Docker images (even slimmed down) have an OS (most likely a Linux distribution), some libs and the runtime needed for a programming language. On it's own a "game engine + WASM binary" is not comparable.
> "Docker images (even slimmed down) have an OS (most likely a Linux distribution)"
Not necessarily. Docker images can be based from the "scratch" container, and may contain only a single binary. With static compilation, the libraries can be included in the binary.
The O/S is provided by the container-host (which is usually a linux VM).
10MB for the Google Homepage! 44MB for Facebook Homepage! I have not been paying attention to website bloat. Wow, and people were annoyed when a site had to download a whole JQuery library for a single function.
It's more or less an abomination. Dropping the frameworks and js jazz you could have most of the features of either of these sites served in the kb range. It would save millions of dollars in electricity, billions in infrastructure, and be more secure.
Yes, it's absolutely the frameworks. That each weigh <100KB. Definitely frameworks :D
Edit:
I don't even know where you lot get these numbers from nowadays, smh.
Just checked, in Slovakia, google.com is 1MB (compressed) total with cache disabled. 400kib of those is my own extension that I installed which is counted among the 'loaded scripts'. Loads in 400ms, blink of an eye.
Framework rants are completely detached from reality, as always.
It's the 35MB wasm, plus the browser runtime, the OS graphical session and the kernel runtime. Docker images are MB because they pack the distro overlay, not because it's unreasonably unoptimized as the clickbaity title suggests.
Docker doesn’t even need a whole OS overlay if your project doesn’t call for it. It’s pretty easy to take a Go app for example and just include it in a scratch container.
> Same as ARM nodes a few years ago: cheaper, denser, widely available – still not the default choice.
Because ARM nodes save you a bit of money (they're cheaper but a bit slower, maybe you end up saving 20-30%) but the move isn't trivial for many tech stacks. When you try it you find you have some Python dependency with a C bit that doesn't have an arm wheel, or your browser automation can't run Chrome on those nodes (there isn't a linux/arm64 build of it). If you're using Go you can cross-compile, although it will likely fail without explicitly disabling cgo, and do you know what the consequences of that are?
Basically it very often ended up being more trouble than it was worth.
I've been playing with Golang and WASM lately; hands-on WASM was new to me.
I found that many dependencies in the ecosystem (especially older ones) do not support GOARCH=wasm nor GOOS=js / GOOS=wasip1. I've had to fork and add support and then do go.mod replace directives. It can get messy.
Golang build tags make it awesome to have different implementations for different systems.
In the browser, it's all single threaded, so goroutines starve each other. I had to put in "breaths" for interactivity.
There's no local filesystem, so you have to figure out other solutions. Some dependencies use the filesystem as an implementation detail or try to shell out. The program will build, but will error at runtime.
That said, it is pretty sweet when it works. You can make WASM games with ebitengine [1] and it emits instructions for a WebGPU renderer; very efficient and many interactivity concerns are handled for you. The NTCharts demo page [2] combines Zig (Ghostty), WASM+Typescript+GLSL (Ghostty Web), and Golang (booba/ntcharts). The WASM size for the demos there is ~5MB each.
My goal is to make tools for terminal remoting and simplify bringing TUIs to the browser.
[1] https://ebitengine.org
Tbh, 35 MBytes for a game client binary isn't much to brag about, WASM or not.
For game engines the underlying problem is usually that the engine is designed around the idea that "everything is a virtual interface" (e.g. 'jump tables') which can't be dead-code eliminated and all engine features are always included, even when not used by a specific game.
I recently ported an old brickout clone I made to Sokol (a C header-based game library). The whole executable is 500kb (macos), surely could be smaller with eg symbols stripped, and it has a whole 3d engine (not that i'm using much more than one custom shader to blit the screen, but it is using 3d engine infrastructure nonetheless). I was impressed that in this day and age such efficiency is still fashionable in some corners. The whole game is about 2mb zipped. Are shameless plugs allowed? If you're curious have a peek! github.com/chrishulbert/brickwarrior
Sure but in WASM, the browser's environment is taking responsibility for many OS concerns (graphics, IO, 3D, hardware). In a Docker image, all that logic needs to be bundled in libraries as you'll only get to "reuse" the hosts' kernel.
A 'fairer' comparison would be a optimized and compiled binary that dynamically links to the OS versus a WASM product (would be kilobytes-megabytes).
Or having the WASM app in a Chromium browser in Docker (would be gigabytes).
WASM sounds great in theory but it's so much pain in practice. Once as a presentation demo I made it run on ESP32 but in the process of digging deep into runtimes and WASM spec, I wanted to disclose pros vs cons and it killed my presentation.
I realize wasm wasn't designed for embedded but it made me open my eyes to it's intricacies like minimum memory allocation, why not native 1 byte variables?
Well my original idea was to show that in a future anyone could code for MCUs in any language. I actually changed my mind in the process. Now we got LLMs and people just prompt anyway.
> Same as ARM nodes a few years ago: cheaper, denser, widely available – still not the default choice.
And rightfully so. Not everything can be guaranteed to work on ARM. It would confuse the hell out of less experienced people trying to launch a simple VPS or something.
Docker's layer-based deduplication mechanism is ineffective or actively harmful IMO. In practice, different images aren't going to share layers even for the same software because images are always going to be built against the latest base image. To make matters worse, many Dockerfiles run `apt-get update` or other equivalent commands, overwriting what's in the base layers and creating tons of wasted space. Nix does a much better job at deduplication with the Nix store.
Also I think Godot's WASM outputs used to be 2-3x smaller in Godot 3 (though the C# one was bigger).
You can ship a JS game in kilobytes, although atop 30 million lines of Chromium that's cheating of course :) still good fun.
https://js13kgames.com/
Notch also made a bunch of 4kb Java (not JS) games back in the day, which is probably what got me into this stuff.
https://web.archive.org/web/20090108001738/http://www.mojang...
https://floooh.github.io/pacman.c/pacman.html
...can probably be made quite a bit smaller when 'as-small-as-possible' is the main goal.
https://github.com/srv1n/kurpod/pkgs/container/kurpod-server
wasm version compiles to under a MB though!
Meanwhile: Error - The following features required to run Godot projects on the Web are missing: WebGL2 - Check web browser configuration and hardware support
This is on Chrome 148.0.7778.96 (Official Build) (64-bit) on Fedora 44, 14900K, 4090 RTX, 128GB RAM
35MB WASM which relies on browser, which relies on drivers and OS, AND it doesn't work as advertised. I get the point, but point was mistaken.
It is true, but besides the point because we have no choice. It is not the 1980s anymore and we can't access the hardware directly on personal computers, vendors don't let us. Not that we would want to, for portability reasons. So we depend on abstraction layers, which are not particularly lightweight, and here, it is a working browser supporting WebGL2 and WASM. You simply can't have a game engine running on modern hardware without a software platform. For the sake of argument, we could imagine a lightweight (tens of MB or less) WASM bare metal platform with simple pixel pushing hardware.
[0] https://caniuse.com/webgl2
https://caniuse.com/webgl2
Only Safari was late to the party (2021).
No offense, but the problem is clearly with your setup.
...tbf, this is mostly a problem on Linux nowadays.
E.g. the MSVC runtime library thing had been solved since Win10 via the "Universal C Runtime".
[1] https://www.construct.net
But the thing with what OP did is, that it runs in the browser and uses a common game engine.
Chrome seems to be ~404MB installed on here; that is conspicuously missing from the comparisons here to Docker containers which do account for more or less the complete runtime.
docker run -it --rm --pid=container:myapp --network=container:myapp --cap-add SYS_PTRACE nicolaka/netshoot
kubectl debug -it pod-name --image=nicolaka/netshoot
The O/S is provided by the container-host (which is usually a linux VM).
This doc describes building an image from "scratch": https://docs.docker.com/build/building/base-images/#create-a...
This site used to have interesting technical discussions, but now its all either AI cheerleading, or vibe coder blog posts.
I did a cache-cleared load of www.google.com just now and didn't get anything close to that.
Edit:
I don't even know where you lot get these numbers from nowadays, smh.
Just checked, in Slovakia, google.com is 1MB (compressed) total with cache disabled. 400kib of those is my own extension that I installed which is counted among the 'loaded scripts'. Loads in 400ms, blink of an eye.
Framework rants are completely detached from reality, as always.
> 10MB for the Google Homepage!
Because ARM nodes save you a bit of money (they're cheaper but a bit slower, maybe you end up saving 20-30%) but the move isn't trivial for many tech stacks. When you try it you find you have some Python dependency with a C bit that doesn't have an arm wheel, or your browser automation can't run Chrome on those nodes (there isn't a linux/arm64 build of it). If you're using Go you can cross-compile, although it will likely fail without explicitly disabling cgo, and do you know what the consequences of that are?
Basically it very often ended up being more trouble than it was worth.
I found that many dependencies in the ecosystem (especially older ones) do not support GOARCH=wasm nor GOOS=js / GOOS=wasip1. I've had to fork and add support and then do go.mod replace directives. It can get messy.
Golang build tags make it awesome to have different implementations for different systems.
In the browser, it's all single threaded, so goroutines starve each other. I had to put in "breaths" for interactivity.
There's no local filesystem, so you have to figure out other solutions. Some dependencies use the filesystem as an implementation detail or try to shell out. The program will build, but will error at runtime.
That said, it is pretty sweet when it works. You can make WASM games with ebitengine [1] and it emits instructions for a WebGPU renderer; very efficient and many interactivity concerns are handled for you. The NTCharts demo page [2] combines Zig (Ghostty), WASM+Typescript+GLSL (Ghostty Web), and Golang (booba/ntcharts). The WASM size for the demos there is ~5MB each.
My goal is to make tools for terminal remoting and simplify bringing TUIs to the browser. [1] https://ebitengine.org
[2] https://nimblemarkets.github.io/ntcharts/demos/heatpicture-p... press 't' for kitty graphics
For game engines the underlying problem is usually that the engine is designed around the idea that "everything is a virtual interface" (e.g. 'jump tables') which can't be dead-code eliminated and all engine features are always included, even when not used by a specific game.
The thing runs in a browser that has most of the stuff implemented to integrate with whatever system it runs on, and just needs to present.
It’s payload versus payload + runtime.
A 'fairer' comparison would be a optimized and compiled binary that dynamically links to the OS versus a WASM product (would be kilobytes-megabytes).
Or having the WASM app in a Chromium browser in Docker (would be gigabytes).
I realize wasm wasn't designed for embedded but it made me open my eyes to it's intricacies like minimum memory allocation, why not native 1 byte variables?
And rightfully so. Not everything can be guaranteed to work on ARM. It would confuse the hell out of less experienced people trying to launch a simple VPS or something.
wtf.
Is there a cleanup tutorial somewhere?