Android Chrome on Pixel 7a: "cropout" just shows 0 % loading bar2, "car configurator" has loading bar go up to 97 or 98 %, but then also doesn't continue.
Sorry to hear that! I will say that usually if you wait long enough, it will eventually load. Try popping open your dev console sidebar, you should see assets downloading over the network.
If it does crash, you'll be able to see why. I'd be interested in seeing any bug reports if you do fine some, we're always squashing bugs over here!
As I see it, the current state of graphics API is worse now than the OpenGL era, despite its promises none of the modern API's are easier to use, truly portable and cross platform.
Having to reinvent OpenGL by creating custom wrappers around Vulkan, Metal, DirectX12, etc is such a time waster as dropping strings and going back to raw char arrays in the name of performance on every modern language.
What promises were made, by whom? Graphics APIs have never been about ease of use as a first order goal. They've been about getting code and data into GPUs as fast as reasonably possible. DevEx will always play second fiddle to that.
I think WebGPU is a decent wrapper for exposing compute and render in the browser. Not perfect by any means - I've had a few paper cuts working with the API so far - but a lot more discoverable and intuitive than I ever found WebGL and OpenGL.
> They've been about getting code and data into GPUs as fast as reasonably possible. DevEx will always play second fiddle to that.
That's a tiny bit revisionist history. Each new major D3D version (at least before D3D12) also fixes usability warts compared to the previous version with D3D11 probably being the most convenient to use 3D API - while also giving excellent performance.
Metal also definitely has a healthy balance between convenience and low overhead - and more recent Metal versions are an excellent example that a high performance modern 3D API doesn't have to be hard to use, nor require thousands of lines of boilerplate to get a triangle on screen.
OTH, OpenGL has been on a steady usability downward trend since the end of the 1990s, and Vulkan unfortunately had continued this trend (but may steer into the right direction in the future:
I hear you but I also don't see a ton of disagreement here either. Like, the fact that D3D12 includes _some_ usability fixes suggests that DevEx really does take a back seat to the primary goal.
I'm not arguing that DevEx doesn't exist in graphics programming. Just that it's second to dots on screen. I also find webgpu to be a lot nicer in terms of DevEx than WebGL.
Wdyt? Still revisionist, or maybe just a slightly different framing of the same pov?
> I also find webgpu to be a lot nicer in terms of DevEx than WebGL.
Amen.
IMHO a new major and breaking D3D version is long overdue. There must be plenty of learnings in which areas it was actually worth it to sacrifice ease-of-use for peformance and where it wasn't.
Or maybe something completely radical/ridiculous and make HLSL the new "D3D API" (with some parts of HLSL code running on the CPU, just enough to prepare CPU side data for upload to the GPU).
> Metal also definitely has a healthy balance between convenience and low overhead - and more recent Metal versions are an excellent example that a high performance modern 3D API doesn't have to be hard to use, nor require thousands of lines of boilerplate to get a triangle on screen.
Metal 4 has moved a lot in the other direction, and now copies a lot of concepts from Vulkan.
WebGPU is both years too late, and just a bit early. Wheras WebGL was OpenGL circa 2005, WebGPU is native graphics circa 2015. It shouldn't need to be said that the bleeding edge new standard for web graphics shouldn't be both 10 years out of date and awful.
Vendors are finally starting to deprecate the old binding model as the byzantine machinery that it is. Bindless resources are an absolute necessity for the modern style of rendering with nanite and raytracing.
Rust's WGPU on native supports some of this, but WebGPU itself doesn't.
It's only intuitive if you don't realize just how huge the gap is between dispatching a vertex shader to render some triangles, and actually producing a lit, shaded and occlusioned image with PBR, indirect lighting, antialiasing and postfx. Would you like to render high quality lines or points? Sorry, it's not been a priority to make that simple. Better go study up on SDFs and beziers.
Which, tbh, is the impression I get from webgpu efforts. Everyone forgets the drivers have been playing pretend for decades, and very few have actually done the homework. Of those that have, most are too enamored with being a l33t gfx coder to realize how terrible the dev exp is.
I'm not sure I disagree with you really - and I ack that webgpu feels like 2015 tech to someone who knows their stuff. I don't have a take on "l33t gfx coder"; I'm a hobbyist not a professional, and I've enjoyed getting up to speed with WebGPU over and above my experiences with WebGL. Happy to be schooled.
I've never impl PBF or raytracing because my interests haven't gone that way. I don't find SDFs to be a particularly difficult concept to "study up on" either though. It's about as close to math-as-drawing that I've seen and doesn't require much more than a couple triangles and a fragment shader. By contrast I've been learning about SVT for a couple months and still haven't quite pieced together a working impl in webgpu... though I understand there are extensions specifically in support of virtual tiling that WebGPU could pursue in a future version.
Agreed DevEx broadly isn't great when working on graphics. But WebGPU feels like a considerable improvement rather than a step backward.
If it weren't for the brand new shading language it might have been a step forward. But instead it's further fragmentation. Vulkan runs happily with GLSL, Proton runs HLSL on Linux, SPIR-V isn't bad.
And the new shading language is so annoying to write it basically has to be generated. Weird shader compilation stuff was already one of the biggest headaches in graphics. Feels like it'll be decades before it'll all be stable.
While I am also not happy with WGSL, note that GLSL has reached a dead end, Khronos officially isn't developing it any further other than extensions, see Vulkanised 2024 talks/panel.
Hence why NVidia's slang offer was welcomed with open arms.
I can give a bit more context as someone that got on WebGL, then WebGPU, and is now picking up Vulkan for the first time.
The problem is that GPU hardware is rapidly changing to enable easier development while still having low level control. With ReBAR for example you can just take a pointer into gigabytes of GPU memory and pump data into it as if it was plain old RAM with barely any performance loss. 100 lines of bullshit suddenly turn into a one line memcpy.
Vulkan is changing to support all this stuff, but the Vulkan API was (a) designed when it didn't exist and is (b) fucking awful. I know that might be a hot take, and I'm still going to use it for serious projects because there's nothing better right now, but the same extensibility that makes it possible for Vulkan to just pivot huge parts of the API to support new stuff also makes it dogshit to use day to day, the code patterns are terrible and it feels like you're constantly compromising on readability at every turn because there is simply zero good options for how to format your code.
WebGPU doesn't have those problems, I quite liked it as an API. But it's based on a snapshot of these other APIs right at the moment before all this work has been done to simplify graphics programming as a whole. And trying to bolt new stuff onto WebGPU in the same way Vulkan is doing is going to end up turning WebGPU into a bloated pile of crap right alongside it.
If you're coming from WebGL, WebGPU is going to feel like an upgrade (or at least it did for me). But now that I've seen a taste of the future I'm pretty sure WebGPU is dead on arrival, it just had horrendous timing, took too long to develop, and now it's backed into a corner. And in the same vein, I don't think extending Vulkan is the way forward, it feels like a pretty big shift is happening right now and IMO that really should involve overhauls at the software/library level too. I don't have experience with DX12 or Metal but I wouldn't be surprised if all 3 go bye bye soon and get replaced with something new that is way simpler to develop with and reflects the current state of hardware and driver capabilities.
>It's only intuitive if you don't realize just how huge the gap is between dispatching a vertex shader to render some triangles, and actually producing a lit, shaded and occlusioned image with PBR, indirect lighting, antialiasing and postfx. Would you like to render high quality lines or points? Sorry, it's not been a priority to make that simple. Better go study up on SDFs and beziers.
I think this is a tad unfair. You're basically describing a semi-robust renderer at that point. IMO to make implementing such a renderer truly "intuitive" (I don't know what this word means to you, so I'm taking it to mean--offloading these features to the API itself) would require railroading the developer some, which appears to go against the design of modern graphics APIs.
I think Unity/Unreal/Godot/Bevy make more sense if you're trying to quickly iterate such features. But even then, you may have to hand write the shader code yourself.
From Steve Wittens, a well respected graphics hacker, and maker of the excellent Use.GPU. https://acko.net/tv/usegpu/ . I'm mostly posting to expand context, and sprinkle in a couple light options.
> Bindless resources are an absolute necessity for the modern style of rendering with nanite and raytracing.
Yeah, for real. Looking at the November 2024 post "What's next for WebGPU" and HN comments, bindless is pretty high up there! There's a high level field survey & very basic proposal (in the hackmd link), and wgpu seems to be filling in the many gaps and seemingly quite far along in implementation. Not seeing any signs yet that the broader WebGPU implementors/spec folks are involved or following along, but at least wgpu is very cross platform & well regarded.
> Would you like to render high quality lines or points? Sorry, it's not been a priority to make that simple. Better go study up on SDFs and beziers.
I realize lines and font rendering are an insanely complex fields, and that OpenGL offering at least lines and Vulkan not sure feels like a slap in the face. The work being done by groups like https://linebender.org/ is intense. Overall though that intensity makes me question the logic of trying to include it, wonders whether fighting to specify something that clearly we don't have full mastery over makes sense: even the very best folks are still improving the craft. We could specify an API without specifying an exact implementation, without conformance tests, perhaps, but that feels like a different risk. Maybe having to reach for a library that does the work reflects where we are, causes the iteration & development we sort of need?
> actually producing a lit, shaded and occlusioned image with PBR, indirect lighting, antialiasing and postfx
I admit to envying the ambition to make this simple, to have such a great deep knowledge as Steve and to think such hard things possible.
I really really am so thankful and hope funding can continue for the incredibly hard work of developing webgpu specs & implementations, and wgpu. As @animats chimes in in the HN submission, bindless in particular is quite a crisis, which either will enable the web to go forward, or remain a lasting real barrier to the web's growth. Really seems to be the tension of Steve's opening position:
> WebGPU is both years too late, and just a bit early. Wheras WebGL was OpenGL circa 2005, WebGPU is native graphics circa 2015.
WebGPU does have line (and point) primitives since they are a direct GPU feature.
It just doesn't bother to 'emulate' lines or points that are wider than 1 pixel, since this is not commonly supported in modern native 3D APIs. Drawing thick lines and points are better done by a high level vector drawing API.
I don't see the problem. There have been lower-level APIs in the graphics stack for a long time (e.g. Mesa's Gallium), only now they are standardised and people are actually choosing to use them. It's not like higher-level APIs don't exist now, OpenGL is still supported on reasonable platforms and WebGPU has been usable from native code for some time.
As for true portability of those low-level APIs, you've basically got Apple to blame (and game console manufacturers, but I don't think anyone expected them to cooperate).
Yeah, that's the thing that really irks me. WebGPU could have been just a light wrapper over Vulkan like WebGL is (or was, it's complicated now) for OpenGL. But apple has been on a dumb war with Khronos for the last decade which has made everything more difficult.
So now we have n+1 low level standards for GPU programming not because we needed them, but because 1 major player is obstinate.
That would be a problem indeed if Metal wouldn't be a much better designed API than Vulkan. As it stands, Vulkan would do good to 'steal' a few ideas from Metal to make the Vulkan API more convenient to use without sacrificing too much performance.
At least for webgpu, you cant wrap around some of the nonse. Like bindigs. They are an inherently bad concept and the way to go would be bindless, but you cant make a wraper that makes an API that revolves around binding into a bindless API. And there are other things like that, and I'm sure there are plenty between Vulkan and Metal.
The problem is, for that to happen Vulkan must be convenient and enjoyable enough to use that people are willing to put their free time into such middleware layers, otherwise it mostly won't happen (unless Valve is willing to sponsor the development - because outside of Valve's effort to make Windows games run on Linux via layering D3D on top of Vulkan, Vulkan is actually quite irrelevant).
How is Apple solely to blame when there are multiple parties involved ? They went to Khronos to turn AMD’s mantle into a true unified next gen APi. Khronos and NVIDIA shot them down to further AZDO OpenGL. Therefore Metal came to be and then DX12 followed and then Vulkan when Khronos realized they had to move that way.
But even if you exclude Metal, what about Microsoft and D3D? Also similarly non-portable. Yet it’s the primary API in use for non-console graphics. You rarely see people complaining about the portability of DX for some reason…
And then in an extremely distant last place is Vulkan. Very few graphics apps actually use Vulkan directly.
People don't complain about DX portability because Windows has first-party support for Vulkan and OpenGL, unlike macOS. Also, since the XBox also uses DirectX, you get two birds with one stone. And third, you aren't forced to use Microsoft hardware to develop for DirectX (these days, you don't even have to use Windows.)
Basically, people are mad that you need to buy Apple hardware, use Apple software (macOS), Apple tooling (Xcode), just to develop graphics code for iOS and macOS. At least you don't also need to use Apple language (Swift) to use Metal, though I don't have any first-hand experience with their C++ bindings so I can't judge if it's a painful experience or not.
Operating systems do not implement graphics APIs for GPUs. These are created by the GPU manufacturer themselves (AMD, Nvidia, etc.). This includes DirectX drivers, both user-space and kernel-space drivers.
Graphics APIs like DirectX and Vulkan are better thought of as (1) a formal specification for GPU behavior, combined with (2) a small runtime. The actual DX/VK drivers are thin shims around a GPU manufacturer's own driver API.
For AMD, the DirectX / Vulkan / OpenGL graphics drivers share a common layer called "PAL" which AMD has open sourced: <https://github.com/GPUOpen-Drivers/pal>
Apple really isn't that different here: they leave the graphics manufacturers to implement their own drivers. Unfortunately, Apple is the sole graphics manufacturer for their OS, and they've chosen to only implement Metal drivers for their GPUs (and a legacy OpenGL driver too).
It's not that big of a deal though, because Vulkan is supported on macOS through the MetalVK project, which wrap the Vulkan API around the Metal API. And projects like vkd3d wrap the DirectX 12 API around the Vulkan API, which is then wrapped around the Metal API. This is how you're able to run Windows games on Mac via the Game Porting Toolkit or CrossOver, btw.
Windows has third party support for Vulkan and OpenGL. It is NOT first party.
It's definitely more convenient than Mac because it is provided by the driver and so you can almost always guarantee they exist, but Microsoft themselves do not provide them. On Mac, for Vulkan you can use MoltenVK which is also third party, and bundle it in the app, though definitely less convenient and less fully featured.
Regarding Xbox, that's a bit of an odd point because you might as well include iOS as a platform at that point which is a bigger gaming platform than Xbox. At least iOS uses the same Metal as Mac, while Xbox does vary in some ways from Windows. Granted, iOS gaming is much more casual oriented but there are some AAA games as well.
Regarding Swift, Metal has always been ObjC first not swift first. The C++ bindings are just for convenience, but you've never been bound to Swift even before they existed. Regarding Xcode, that's only to get the toolchain or if you need instrumentation. You don't need to use Xcode to actually develop things, this is no more a burden than needing Visual Studio on Windows.
Windows doesn't have first party support for OpenGL and Vulkan.
It has a plugglable driver system, leftover from the Windows NT/OpenGL 1.1 days called ICD, that driver vendors use to add their OpenGL and Vulkan drivers.
I think what we learned from the OpenGL era is it's actually not very relevant all platforms use the same high level API to talk to the GPU hardware. What matters it the platform's chosen API offers good control of the hardware it uses.
You say this requires reinvention but really the end work is "translate OpenGL to something the hardware can actually understand" in both scenarios. The difference with the OpenGL era is you did not have the option to avoid using the wrapper, not that no wrapper existed. Targeting the best of each possibly hardware type individually without baking in assumptions about the hardware has proven to not be very practical, but it only matters if you're building a "easy translation layer" rather than using it or trying to target specific types of hardware very directly (in which case you don't want something super generic or simple, you want something which exposes the hardware as directly as is reasonable for that hardware type).
OpenGL became a mess of an API after 2.0, and WebGPU is actually a fairly easy to use wrapper around Vk, D3D12, Metal - definitely better designed than what OpenGL has become.
Apart from that, D3D11 and Metalv1 are probably the sweet spot between ease-of-use and performance (especially D3D11's performance is hard to beat even in Vulkan and D3D12).
D3D11 is so nice to use that I feel like the ideal workflow for ground-up graphics applications is to just write everything in D3D11 and then let Middleware layers on Linux (proton) or Mac (Game porting toolkit) handle the translation. DirectX also has a whole suite of first party software (DirectXMath, DirectXTK) which make a lot of common workflows much simpler.
If only the windows team could get out of a tailspin because almost everything else MS produces on the Windows side gets worse and worse every year.
I agree. I'll keep using OpenGL with CUDA interop until something better shows up. Vulkan isn't it. I tried Vulkan to get away from OpenGL, but ended up with CUDA instead since it's so much nicer to work with. Vulkan has way too much overengineered complexity with zero benefit.
OpenGL is still such a powerful technology. I use it all the time because Vulkan is just so much more difficult to use. It's a pity, so much good software not being built because ogl is more or less a dead man walking
Not really, for example in the OpenGL era there was this urban myth that game consoles used OpenGL, this was never really the case.
Nintendo after graduating to devkits where C and C++ could be used like N64, had OpenGL inspired APIs, which isn't really the same. Although there was some GLSL like shader support.
They only started supporting Khronos APIs with the Switch, and even then, if you want the full power of the Switch, NVN is the way to go.
Playstation always had proprietary APIs, they did a small stint OpenGL ES 1.0 + Cg, which had very little to no uptake among developers, and they dropped it from the devkits.
Sega only had proprietary APIs, and there was a small collaboration with Microsoft for DirectX, which only a few studios took advantage of.
XBox naturally has always been about DirectX.
Go watch GDC Vault programming track to see how many developers you will find complaining about writing middleware for their game engines, if any at all, versus how many talks about taking the advantage of every little low level detail of hardware architecture.
Early console APIs were more similar to Direct3D 1, with very rudimentary immediate mode commands. Modern console APIs still have a less stateful, easy API layer, like D3D10/11, but also expose more low-level stuff, too.
OpenGL didn't match the hardware well except on SGI hardware or carryover descendants like 3dfx.
The point of the graphics APIs is to be as close to the metal as possible. It’s a balancing act between portability and hardware design/performance. I really don’t think it’s as trivial as non-graphics engineers make it out to be to make something universal.
But even when it existed in the form of OpenGL , or now WebGPU, people complain about the performance overhead. So you end up back here.
Vulkan isnt close to the metal, though. It's a high-level wrapper around all the quirks and differences of ancient mobile to modern desktop GPUs. Render passes, for example, are entirely irrelevant for desktop GPUs. They are not close to metal, but add needless complexity. Recently, a Vulkan driver engineer even told me that they are not necessary for tile-based mobile GPUs for which they were intended, since they can figure the necessary things out by themselves. And I would guess they need to, since they became optional in Vulkan, so they cant relly on them anymore. They are still mandatory in WebGPU, for no good reason.
And there are so many pointless things that are no longer relevant, or should at best be optional so that devs can get things done before optimizing.
Yes they’re abstractions, because nobody really wants anyone to be writing directly against the ISA either since the vendors need the ability to change things over time.
Again, to my point, it’s about balancing portability and power/perf.
Yet they ended up creating something that makes OpenGL still an attractive choice. That excessive complexity certainly wasnt necessary.
Personally, I'll sit this generation out and wait dor whatever comes after. I ended up switching to doing software rasterization in Cuda because that's easier than drawing a triangle in Vulkan. Cuda has shown me how insane Vulkan is. Like, why even have descriptir sets, bindings, etc? In cuda you simply call a kernel and provide the data (e.g. vertex or storage buffer) as a pointer argument.
What would a leader do? Nvidia wants to sell hardware, Nintendo wants to sell games, Microsoft wants to either buy Linux or crush it. Nobody has a stake in things actually working
Modern graphics APIs are the graphical equivalent to assembly language - you are not supposed to use them directly, but through another higher level layer, like a programming language or a graphics engine.
They are a specialized API intended for tool writers.
I'm still hoping that WebGPU somehow takes off for non-web use so that we have an easy to use cross platform API with an official spec (a replacement for opengl). However, it seems that outside of the Rust world, there doesn't seem to be much interest for using WebGPU for native code. I don't know any big projects using Dawn for example. Part of the reason seems to be that WebGPU came a bit too late and everyone was already using custom-built abstractions over dx, vulkan and metal.
It won't. It's barely simpler but lacks a lot of functionality. Some stuff that became optional in Vulkan (render passes) are still mandatory on WebGPU, and bind groups are static and thus cumbersome. It also adds additional limitations and cruft, like you can't easily transfer from host to a buffer subregion and need staging buffers.
I'll use it for web since there is no alternative, but for desktop I'll stick with an OpenGL+CUDA interop framework until a sane, modern graphics API shows up. I.e., a graphics API that gets rid of render pases, static pipelines, mandatory explizit syncing, bindings and descriptor sets (simply use buffers and pointers), and all the other nonsense.
If allocating and populating a buffer takes more effort than a simple cuMemAlloc and cuMemcpy, and calling a shader with arguments takes more than simply passing the shader pointers to the data, then I'm out.
...that's assuming that the WebGPU API is set in stone, which hopefully it isn't.
They'd do well to follow the D3D model (major breaking versions, while guaranteeing backward compatibility for older versions) - e.g. WebGPU2, WebGPU3, WebGPU4 each being a mostly new API without having to compromise for backward compatibility.
WebGPU had more features and capabilites back in 2020 before they started removing and limiting them. Forgive my lack of enthusiasm and optimism for the future prospects of an API that was already ancient when development started, and even less capable by the time it was released.
> WebGPU had more features and capabilites back in 2020 before they started removing and limiting them
I think that's the price to pay for trying to cover a wide range of hardware. You can't just make all those shitty Android phones disappear. At least for each WebGPU limit, there's usually a Github ticket which explains why exactly this limit exists.
Yeah, unfortunately I'm in real-time rendering research so I like to play with fairly modern desktop GPUs. The no-phone-left-behind policy made WebGPU a somewhat unattractive target for me. Which is unfortunate because during the early days it felt like we'd get a cutting-edge modern API for the browser and I was excited and ready to abandon OpenGL for WebGPU. Instead, I ended up switching to CUDA which I avoided for years due to platform dependency. But once I noticed how pleasant it is to work with I could not go back to graphics APIs. I really like the "easy things should be easy, complex things should be possible" design of Cuda.
Another reason may be that WebGPU didn't allow for as much optimization and control as Vulkan, and the performance isn't as good as Vulkan. WebGPU also doesn't have all the extensions that Vulkan has.
Even webgl don't give any gurentee about your code will run well on any devices though. It only gurentee that it will run, but it can have 10x performance difference on different platform depends on how you wrote the shaders.
As anyone used to Khronos APIs is aware, that is of little value without actual Conformance Tests Suites, and even then there are plently of forgotten guarantees when using consumer hardware with all the usual OEM quality practices.
Good then that even WebGL mostly doesn't run on Khronos APIs ;) (only on Linux, although I don't know whether ANGLE is now actually using a Vulkan backend on Linux - which of course is also a Khronos API though).
Both WebGL2 and WebGPU are probably the most 'watertight' specced and tested 3D API ever built, and especially WebGPU has gone to great lengths to eliminate UB present in native APIs (even at the cost of usability).
The Android OS and the entire Android ecosystem being a huge pile of excrement isn't really surprising though. But it's 'too big to ignore' unfortunately, at least for browser APIs.
I think the parent is implying there are 1001 soc out there with some form of embedded gpu that probably have issues actually implementing webgpu. Like those in millions of Chinese tablets. Are they likely targets? Probably not now but in 5 years? Mainstream desktop hardware? No problem.
Some part of it is also probably the atrocious naming. I don't do anything with web, only native coding, so whenever I heard something about web gpu somewhere I just ignored it, for literally years, because I just assumed it was some new web tech and thus not relevant to me at all.
Very happy to see this as it means that our gpu-allocator [0] crate (used currently by wgpu's dx12 backend, but capable of supporting vulkan & metal as well) will see a significant wider audience then what we've been using it for so far (which is shipping our gpu benchmark suite: evolve [1]
I have been using wgpu for my main projects for nearly two years now. Let's hope this rollout means more maintainers so issues I have opened 18 months ago bug more people and eventually get resolved. Never touched rust myself but maybe I find the motivation and time to do it myself.
As I also depend on the wgpu-native bindings it's slow for updates to reach. Like we just got to v25 last week and v26 dropped a couple days prior to that.
It runs the SmolLM2 model compiled to WebAssembly for structured data extraction. I previously thought that demo only worked in Chrome.
(If I try it in regular Firefox for Mac I get "Error: WebGPU is not supported in your current environment, but it is necessary to run the WebLLM engine.")
>we plan to ship WebGPU on Mac and Linux in the coming months, and finally on Android
Sounds good. I'm not really thrilled about it as of now. What ever the reason, it's not been supported in Linux for any browsers as of yet. My guess is it's too hard to expose without creating terrible attack surfaces.
This seems to support my view that web standards are too overgrown for how users actually use the web. It's obviously too late to do anything about it now but all the issues of monoculture and funding we are worried about today stem from the complexity of making a web browser due to decisions tracing all the way back to the days of Netscape.
The Keijiro examples work flawlessly on iOS with WebGPU feature flag enabled. Big caveat, but it shows why this stuff matters: Linux desktop, a little, iOS mobile Safari, a lot.
Most of the sites I’ve seen are indeed just demos. I especially like Compute Toys [0], a Shadertoy clone for WebGPU. [1] is probably the best place to find demos. I have a site myself in which I experiment mainly with WebGPU Compute Shaders [2].
Compute shaders, which can draw points faster than the native rendering pipeline. Although I have to admit that WebGPU implements things so poorly and restrictive, that this benefit ends up being fairly small. Storage buffers, which come along with compute shaders, are still fantastic from a dev convenience point of view since it allows implementing vertex pulling, which is much nicer to work with than vertex buffers.
For gaussian splatting, WebGPU is great since it allows implementing sorting via compute shaders. WebGL-based implementations sort on the CPU, which means "correct" front-to-back blending lags behind for a few frames.
But yeah, when you ask like that, it would have been much better if they had simply added compute shaders to WebGL, because other than that there really is no point in WebGPU.
Access to slightly more recent GPU features (e.g. WebGL2 is stuck on a feature set that was mainstream ca. 2008, while WebGPU is on a feature set that was mainstream ca 2015-ish).
The GL programming only feels 'natural' if you've been following GL development closely since the late 1990s and learned to accept all the design compromises for sake of backward compatibility. If you come from other 3D APIs and never touched GL before it's one "WTF were they thinking" after another (just look at VAOs as an example of a really poorly designed GL feature).
While I would have designed a few things differently in WebGPU (especially around the binding model), it's still a much better API than WebGL2 from every angle.
The limited feature set of WebGPU is mostly to blame on Vulkan 1.0 drivers on Android devices I guess, but there's no realistic way to design a web 3D API and ignore shitty Android phones unfortunately.
It's not about feeling natural - I fully agree that OpenGL is a terrible and outdated API. It's about the complete overengengineered and pointless complexity in Vulkan-like APIs and WebGPU. Render Passes are entirely pointless complexity that should not exist. It's even optional in Vulkan nowadays, but still mandatory in WebGPU. Similarly static binding groups are entirely pointless, now I've got to cache thousands of vertex and storage buffers. In Vulkan you can nowadays modify those, but not in WebGPU. Wish I could batch them buffers in a single one so I dont need to create thousands of bind groups, but that's also made needlessly cumbersome in WebGPU due to the requirement to use staging buffers. And since buffer sizes are fairly limited, I can't just create one that fits all, so I have to create multiple buffes anyway, might as well have a separate buffer for all nodes. Virtual/Sparse buffers would be helpful in single-buffer designs by growing those as much as needed, but of course they also dont exist in WebGPU.
The one thing that WebGPU is doing better is that it does implicit syncing by default. The problem is, it provides no options for explicit syncing.
I mainly software-rasterize everything in Cuda nowadays, which makes the complexity of graphics apis appear insane. Cuda allows you to get things done simple and easily, but it still has all the functionaility to make things fast and powerful. The important part is that the latter is optinal, so you can get things done quickly, and still make them fast.
In cuda, allocating a buffer and filling it with data is a simple cuMemAlloc and cuMemcpy. When calling a shader/kernel, I dont need bindings and descriptors, I simply pass a pointer to the data. Why would I need that anyway, the shader/kernel knows all about the data, the host doesnt need to know.
> Render Passes are entirely pointless complexity that should not exist. It's even optional in Vulkan nowadays.
AFAIK Vulkan only eliminated pre-baked render pass objects (which were indeed pointless), and now simply copied Metal's design of transient render passes, e.g. there's still 'render pass boundaries' between vkCmdBeginRendering() and vkCmdEndRendering() and the VkRenderingInfo struct that's passed into the vkCmdBeginRendering() function (https://registry.khronos.org/vulkan/specs/latest/man/html/Vk...) is equivalent with Metal's MTLRenderPassDescriptor (https://developer.apple.com/documentation/metal/mtlrenderpas...).
E.g. even modern Vulkan still has render passes, they just didn't want to call those new functions 'Begin/EndRenderPass' for some reason ;) AFAIK the idea of render pass boundaries is quite essential for tiler GPUs.
WebGPU pretty much tries to copy Metal's render pass approach as much as possible (e.g. it doesn't have pre-baked pass objects like Vulkan 1.0).
> The one thing that WebGPU is doing better is that it does implicit syncing by default.
AFAIK also mostly thanks to the 'transient render pass model'.
> Why would I need that anyway, the shader/kernel knows all about the data, the host doesnt need to know.
Because old GPUs are a thing and those usually don't have such a flexible hardware design to make rasterizing (or even vertex pulling) in compute shaders performant enough to compete with the traditional render pipeline.
> Similarly static binding groups are entirely pointless
I agree, but AFAIK Vulkan's 1.0 descriptor model is mostly to blame for the inflexible BindGroups design.
> but that's also made needlessly cumbersome in WebGPU due to the requirement to use staging buffers
Most modern 3D APIs also switched to staging buffers though, and I guess there's not much choice if you don't have unified memory.
> AFAIK the idea of render pass boundaries is quite essential for tiler GPUs.
I've been told by a driver dev of a tiler GPU that they are, in fact, not essential. They pick that info up by themselves by analyzing the command buffer.
> Most modern 3D APIs also switched to staging buffers though, and I guess there's not much choice if you don't have unified memory.
Well I wouldn't know since I switched to using Cuda as a graphics API. It's mostly nonsense-free, and faster than the hardware pipeline for points, and about as fast for splats. Seeing how Nanite also software-rasterizes as a performance improvement, Cuda may even be great for triangles. Only implemented a rudimentary triangle rasterizer that can draw 10 million small textured triangles per millisecond. Still working on the larger ones, but low-priority since I focus on point clouds.
In any case, I won't touch graphics APIs anymore until they make a clean break to remove the legacy nonsense. Allocating buffers should be a single line, providing data to shaders should be as simple as passing pointers, etc..
I keep waiting to see ambitious webgames that could match the experience of Infinity Blade from 2010, used to demo iOS new OpenGL ES 3.0 capabilities, the foundation of WebGL 2.0.
The only thing I like in Web 3D APIs, is that outside middleware engines, they are the only mainstream 3D APIs designed with managed languages in mind, instead of after the fact bindings.
Still waiting for something like RenderDoc on the respective browser developer tools, we never got anything better than SpectorJS.
It isn't even printf debugging, rather pixel colour debugging.
There's no way you'll see anything like that. Flash was dead simple, a 12 year old could throw a simple game together and upload it. WebGPU will require a skilled graphics programmer just to write (or more likely cross compile) these weird shaders.
And the SWF format had insane compatibility, literally unmatched by any other technology imo, we didn't even think about OS's, it really was "write once run anywhere" (pre-smartphone ofc). On the web, even basic CSS doesn't work the same from OS to OS, and WebGL apps still crash on 10% of devices randomly. It'll probably be 5 years before WebGPU is even remotely stable.
Not even to mention the fully integrated editor environment.
Or I guess maybe you're saying someone should build something like Flash targeting WebGPU? Probably the closest there is to that right now is Figma? But it feels weak too imo, and was already possible with WebGL. Maybe Unreal Engine is the bet.
If you follow things like three.js you'll be painfully aware that in truth there doesn't seem to be much use for this at all. "3D on the web" is something that sounds fun until it's possible at which point it becomes meh[1]. The exception proving the rule would be that Marble Madness promo game https://news.ycombinator.com/item?id=42212644
Consequently much of the JS 3D community has become obsessed with gaussian splatting, and AR more generally.
[1] And I would extend this to what's going on here: people prefer complaining about how missing features in APIs prevent their genius idea from being possible, when in truth there's simply no demand from users for this stuff at all. You could absolutely have done web Minecraft years ago, and it's very revealing such a thing is not wildly popular. I personally wasted too long on WebGL ( https://www.luduxia.com/ ), and what I learned is the moment it all works people just assume it was nothing and move on.
There are so many blockers, versus old style Flash games.
Driver and OS blacklisting, means that game developers aren't aware of the user experience, nor can they controll it, as in native games, or server side rendering with streaming.
No proper debugging tools other than printf/pixel debugging.
The amount of loading screens that would be needed, given memory constraints of browser sessions.
This alone means there is hardly that much ROI for 3D webgames, and most uses end up being in ecommerce, or Google Maps kind of applications.
Very cool! Now, let see how much will take for G-products to actually use it and not complaining about "browser not supported for this feature, use Chrome".
The one I can think of is Google Meet, where some GPU-thing is used to add background effects such as blur. However I'm not sure this actually uses WebGPU; it used to use on Firefox until Google added a browser check, and AFAIK, if you could fool Meet to think Firefox was Chrome, it would still work.
This might still be a semi-legitimate thing, i.e maybe they kept around a WebGL implementation for a while as a fallback but moved the main implementation to WebGPU and don't want to maintain the fallback. It certainly fits well into their strategy of making sure that the web really only works properly with Chrome.
Huh really? That must be a relatively recent change, last time I was in a Meet meeting in Firefox I was just blocked by a message saying my browser isn't supported. That is admittedly some time ago though.
I'm on Brave Linux which requires special flag for WebGPU (not turned on for me) and can confirm that background blur still works with Meet without WebGPU
I don't know :) I was referring basically to the Meet situation back in the day with FF where the feature was there but Meet complained the browser was not capable.
There was a genuine technical reason for that, a part of WebRTC that Firefox hadn’t implemented yet, where if even a single member of a group call lacked that feature, it had to fall back to something that used a lot more CPU for everyone. Can’t remember the details exactly, but it was approximately that.
If I remember correctly, the issue was related to newer APIs like MediaStreamTrackProcessor, offscreen surfaces, and WebRTC–WebCodecs interoperability, as well as the ability to run ML inference efficiently in the browser. At the time, Firefox hadn’t fully implemented some of these features, which impacted Google Meet’s ability to apply effects like background blur or leverage hardware-accelerated video processing.
My company is working to bring Unreal to the browser, and we've built out a custom WebGPU RHI for Unreal Engine 5.
Here are demos of the tech in action, for anyone interested:
(Will only work on Chromium-based browsers on desktop, and on some Android phones)
Cropout: https://play-dev.simplystream.com/?token=aa91857c-ab14-4c24-...
Car configurator: https://garage.cjponyparts.com/
This post is about WebGPU in Firefox. Do you plan to test and/or release a Firefox-compatible version?
If it does crash, you'll be able to see why. I'd be interested in seeing any bug reports if you do fine some, we're always squashing bugs over here!
https://topdown.tiwsamples.com/
I think WebGPU is a decent wrapper for exposing compute and render in the browser. Not perfect by any means - I've had a few paper cuts working with the API so far - but a lot more discoverable and intuitive than I ever found WebGL and OpenGL.
That's a tiny bit revisionist history. Each new major D3D version (at least before D3D12) also fixes usability warts compared to the previous version with D3D11 probably being the most convenient to use 3D API - while also giving excellent performance.
Metal also definitely has a healthy balance between convenience and low overhead - and more recent Metal versions are an excellent example that a high performance modern 3D API doesn't have to be hard to use, nor require thousands of lines of boilerplate to get a triangle on screen.
OTH, OpenGL has been on a steady usability downward trend since the end of the 1990s, and Vulkan unfortunately had continued this trend (but may steer into the right direction in the future:
https://www.youtube.com/watch?v=NM-SzTHAKGo
I'm not arguing that DevEx doesn't exist in graphics programming. Just that it's second to dots on screen. I also find webgpu to be a lot nicer in terms of DevEx than WebGL.
Wdyt? Still revisionist, or maybe just a slightly different framing of the same pov?
Amen.
IMHO a new major and breaking D3D version is long overdue. There must be plenty of learnings in which areas it was actually worth it to sacrifice ease-of-use for peformance and where it wasn't.
Or maybe something completely radical/ridiculous and make HLSL the new "D3D API" (with some parts of HLSL code running on the CPU, just enough to prepare CPU side data for upload to the GPU).
Metal 4 has moved a lot in the other direction, and now copies a lot of concepts from Vulkan.
https://developer.apple.com/documentation/metal/understandin...
https://developer.apple.com/documentation/metal/resource-syn...
Technically true, but practically tone deaf.
WebGPU is both years too late, and just a bit early. Wheras WebGL was OpenGL circa 2005, WebGPU is native graphics circa 2015. It shouldn't need to be said that the bleeding edge new standard for web graphics shouldn't be both 10 years out of date and awful.
Vendors are finally starting to deprecate the old binding model as the byzantine machinery that it is. Bindless resources are an absolute necessity for the modern style of rendering with nanite and raytracing.
Rust's WGPU on native supports some of this, but WebGPU itself doesn't.
It's only intuitive if you don't realize just how huge the gap is between dispatching a vertex shader to render some triangles, and actually producing a lit, shaded and occlusioned image with PBR, indirect lighting, antialiasing and postfx. Would you like to render high quality lines or points? Sorry, it's not been a priority to make that simple. Better go study up on SDFs and beziers.
Which, tbh, is the impression I get from webgpu efforts. Everyone forgets the drivers have been playing pretend for decades, and very few have actually done the homework. Of those that have, most are too enamored with being a l33t gfx coder to realize how terrible the dev exp is.
I've never impl PBF or raytracing because my interests haven't gone that way. I don't find SDFs to be a particularly difficult concept to "study up on" either though. It's about as close to math-as-drawing that I've seen and doesn't require much more than a couple triangles and a fragment shader. By contrast I've been learning about SVT for a couple months and still haven't quite pieced together a working impl in webgpu... though I understand there are extensions specifically in support of virtual tiling that WebGPU could pursue in a future version.
Agreed DevEx broadly isn't great when working on graphics. But WebGPU feels like a considerable improvement rather than a step backward.
And the new shading language is so annoying to write it basically has to be generated. Weird shader compilation stuff was already one of the biggest headaches in graphics. Feels like it'll be decades before it'll all be stable.
Hence why NVidia's slang offer was welcomed with open arms.
The problem is that GPU hardware is rapidly changing to enable easier development while still having low level control. With ReBAR for example you can just take a pointer into gigabytes of GPU memory and pump data into it as if it was plain old RAM with barely any performance loss. 100 lines of bullshit suddenly turn into a one line memcpy.
Vulkan is changing to support all this stuff, but the Vulkan API was (a) designed when it didn't exist and is (b) fucking awful. I know that might be a hot take, and I'm still going to use it for serious projects because there's nothing better right now, but the same extensibility that makes it possible for Vulkan to just pivot huge parts of the API to support new stuff also makes it dogshit to use day to day, the code patterns are terrible and it feels like you're constantly compromising on readability at every turn because there is simply zero good options for how to format your code.
WebGPU doesn't have those problems, I quite liked it as an API. But it's based on a snapshot of these other APIs right at the moment before all this work has been done to simplify graphics programming as a whole. And trying to bolt new stuff onto WebGPU in the same way Vulkan is doing is going to end up turning WebGPU into a bloated pile of crap right alongside it.
If you're coming from WebGL, WebGPU is going to feel like an upgrade (or at least it did for me). But now that I've seen a taste of the future I'm pretty sure WebGPU is dead on arrival, it just had horrendous timing, took too long to develop, and now it's backed into a corner. And in the same vein, I don't think extending Vulkan is the way forward, it feels like a pretty big shift is happening right now and IMO that really should involve overhauls at the software/library level too. I don't have experience with DX12 or Metal but I wouldn't be surprised if all 3 go bye bye soon and get replaced with something new that is way simpler to develop with and reflects the current state of hardware and driver capabilities.
I think this is a tad unfair. You're basically describing a semi-robust renderer at that point. IMO to make implementing such a renderer truly "intuitive" (I don't know what this word means to you, so I'm taking it to mean--offloading these features to the API itself) would require railroading the developer some, which appears to go against the design of modern graphics APIs.
I think Unity/Unreal/Godot/Bevy make more sense if you're trying to quickly iterate such features. But even then, you may have to hand write the shader code yourself.
> Bindless resources are an absolute necessity for the modern style of rendering with nanite and raytracing.
Yeah, for real. Looking at the November 2024 post "What's next for WebGPU" and HN comments, bindless is pretty high up there! There's a high level field survey & very basic proposal (in the hackmd link), and wgpu seems to be filling in the many gaps and seemingly quite far along in implementation. Not seeing any signs yet that the broader WebGPU implementors/spec folks are involved or following along, but at least wgpu is very cross platform & well regarded.
https://developer.chrome.com/blog/next-for-webgpu https://news.ycombinator.com/item?id=42209272 https://hackmd.io/PCwnjLyVSqmLfTRSqH0viA https://hackmd.io/@cwfitzgerald/wgpu-bindless https://github.com/gfx-rs/wgpu/issues/3637 https://github.com/gpuweb/gpuweb/issues/380
> Would you like to render high quality lines or points? Sorry, it's not been a priority to make that simple. Better go study up on SDFs and beziers.
I realize lines and font rendering are an insanely complex fields, and that OpenGL offering at least lines and Vulkan not sure feels like a slap in the face. The work being done by groups like https://linebender.org/ is intense. Overall though that intensity makes me question the logic of trying to include it, wonders whether fighting to specify something that clearly we don't have full mastery over makes sense: even the very best folks are still improving the craft. We could specify an API without specifying an exact implementation, without conformance tests, perhaps, but that feels like a different risk. Maybe having to reach for a library that does the work reflects where we are, causes the iteration & development we sort of need?
> actually producing a lit, shaded and occlusioned image with PBR, indirect lighting, antialiasing and postfx
I admit to envying the ambition to make this simple, to have such a great deep knowledge as Steve and to think such hard things possible.
I really really am so thankful and hope funding can continue for the incredibly hard work of developing webgpu specs & implementations, and wgpu. As @animats chimes in in the HN submission, bindless in particular is quite a crisis, which either will enable the web to go forward, or remain a lasting real barrier to the web's growth. Really seems to be the tension of Steve's opening position:
> WebGPU is both years too late, and just a bit early. Wheras WebGL was OpenGL circa 2005, WebGPU is native graphics circa 2015.
WebGPU does have line (and point) primitives since they are a direct GPU feature.
It just doesn't bother to 'emulate' lines or points that are wider than 1 pixel, since this is not commonly supported in modern native 3D APIs. Drawing thick lines and points are better done by a high level vector drawing API.
As for true portability of those low-level APIs, you've basically got Apple to blame (and game console manufacturers, but I don't think anyone expected them to cooperate).
Yeah, that's the thing that really irks me. WebGPU could have been just a light wrapper over Vulkan like WebGL is (or was, it's complicated now) for OpenGL. But apple has been on a dumb war with Khronos for the last decade which has made everything more difficult.
So now we have n+1 low level standards for GPU programming not because we needed them, but because 1 major player is obstinate.
Being simpler is an advantage. It means that 3rd party GPU drivers can more simply implement the interface correctly.
However, as discussed in other comments, that doesn't change the driver quality mess of the platform.
How is Apple solely to blame when there are multiple parties involved ? They went to Khronos to turn AMD’s mantle into a true unified next gen APi. Khronos and NVIDIA shot them down to further AZDO OpenGL. Therefore Metal came to be and then DX12 followed and then Vulkan when Khronos realized they had to move that way.
But even if you exclude Metal, what about Microsoft and D3D? Also similarly non-portable. Yet it’s the primary API in use for non-console graphics. You rarely see people complaining about the portability of DX for some reason…
And then in an extremely distant last place is Vulkan. Very few graphics apps actually use Vulkan directly.
Have you tried writing any of the graphics APIs?
Basically, people are mad that you need to buy Apple hardware, use Apple software (macOS), Apple tooling (Xcode), just to develop graphics code for iOS and macOS. At least you don't also need to use Apple language (Swift) to use Metal, though I don't have any first-hand experience with their C++ bindings so I can't judge if it's a painful experience or not.
Operating systems do not implement graphics APIs for GPUs. These are created by the GPU manufacturer themselves (AMD, Nvidia, etc.). This includes DirectX drivers, both user-space and kernel-space drivers.
Graphics APIs like DirectX and Vulkan are better thought of as (1) a formal specification for GPU behavior, combined with (2) a small runtime. The actual DX/VK drivers are thin shims around a GPU manufacturer's own driver API.
For AMD, the DirectX / Vulkan / OpenGL graphics drivers share a common layer called "PAL" which AMD has open sourced: <https://github.com/GPUOpen-Drivers/pal>
Apple really isn't that different here: they leave the graphics manufacturers to implement their own drivers. Unfortunately, Apple is the sole graphics manufacturer for their OS, and they've chosen to only implement Metal drivers for their GPUs (and a legacy OpenGL driver too).
It's not that big of a deal though, because Vulkan is supported on macOS through the MetalVK project, which wrap the Vulkan API around the Metal API. And projects like vkd3d wrap the DirectX 12 API around the Vulkan API, which is then wrapped around the Metal API. This is how you're able to run Windows games on Mac via the Game Porting Toolkit or CrossOver, btw.
It's definitely more convenient than Mac because it is provided by the driver and so you can almost always guarantee they exist, but Microsoft themselves do not provide them. On Mac, for Vulkan you can use MoltenVK which is also third party, and bundle it in the app, though definitely less convenient and less fully featured.
Regarding Xbox, that's a bit of an odd point because you might as well include iOS as a platform at that point which is a bigger gaming platform than Xbox. At least iOS uses the same Metal as Mac, while Xbox does vary in some ways from Windows. Granted, iOS gaming is much more casual oriented but there are some AAA games as well.
Regarding Swift, Metal has always been ObjC first not swift first. The C++ bindings are just for convenience, but you've never been bound to Swift even before they existed. Regarding Xcode, that's only to get the toolchain or if you need instrumentation. You don't need to use Xcode to actually develop things, this is no more a burden than needing Visual Studio on Windows.
It has a plugglable driver system, leftover from the Windows NT/OpenGL 1.1 days called ICD, that driver vendors use to add their OpenGL and Vulkan drivers.
https://learn.microsoft.com/en-us/windows-hardware/drivers/d...
In some subsystems like UWP, or Windows on ARM, ICDs aren't supported, and OpenGL/Vulkan have to be mapped on top of DirectX.
https://devblogs.microsoft.com/directx/announcing-the-opencl...
You say this requires reinvention but really the end work is "translate OpenGL to something the hardware can actually understand" in both scenarios. The difference with the OpenGL era is you did not have the option to avoid using the wrapper, not that no wrapper existed. Targeting the best of each possibly hardware type individually without baking in assumptions about the hardware has proven to not be very practical, but it only matters if you're building a "easy translation layer" rather than using it or trying to target specific types of hardware very directly (in which case you don't want something super generic or simple, you want something which exposes the hardware as directly as is reasonable for that hardware type).
Apart from that, D3D11 and Metalv1 are probably the sweet spot between ease-of-use and performance (especially D3D11's performance is hard to beat even in Vulkan and D3D12).
If only the windows team could get out of a tailspin because almost everything else MS produces on the Windows side gets worse and worse every year.
Nintendo after graduating to devkits where C and C++ could be used like N64, had OpenGL inspired APIs, which isn't really the same. Although there was some GLSL like shader support.
They only started supporting Khronos APIs with the Switch, and even then, if you want the full power of the Switch, NVN is the way to go.
Playstation always had proprietary APIs, they did a small stint OpenGL ES 1.0 + Cg, which had very little to no uptake among developers, and they dropped it from the devkits.
Sega only had proprietary APIs, and there was a small collaboration with Microsoft for DirectX, which only a few studios took advantage of.
XBox naturally has always been about DirectX.
Go watch GDC Vault programming track to see how many developers you will find complaining about writing middleware for their game engines, if any at all, versus how many talks about taking the advantage of every little low level detail of hardware architecture.
OpenGL didn't match the hardware well except on SGI hardware or carryover descendants like 3dfx.
There are no adults, no leaders with an eye on things leading us away from further mistakes, and we keep going deeper.
But even when it existed in the form of OpenGL , or now WebGPU, people complain about the performance overhead. So you end up back here.
And there are so many pointless things that are no longer relevant, or should at best be optional so that devs can get things done before optimizing.
Yes they’re abstractions, because nobody really wants anyone to be writing directly against the ISA either since the vendors need the ability to change things over time.
Again, to my point, it’s about balancing portability and power/perf.
Personally, I'll sit this generation out and wait dor whatever comes after. I ended up switching to doing software rasterization in Cuda because that's easier than drawing a triangle in Vulkan. Cuda has shown me how insane Vulkan is. Like, why even have descriptir sets, bindings, etc? In cuda you simply call a kernel and provide the data (e.g. vertex or storage buffer) as a pointer argument.
They are a specialized API intended for tool writers.
I'll use it for web since there is no alternative, but for desktop I'll stick with an OpenGL+CUDA interop framework until a sane, modern graphics API shows up. I.e., a graphics API that gets rid of render pases, static pipelines, mandatory explizit syncing, bindings and descriptor sets (simply use buffers and pointers), and all the other nonsense.
If allocating and populating a buffer takes more effort than a simple cuMemAlloc and cuMemcpy, and calling a shader with arguments takes more than simply passing the shader pointers to the data, then I'm out.
They'd do well to follow the D3D model (major breaking versions, while guaranteeing backward compatibility for older versions) - e.g. WebGPU2, WebGPU3, WebGPU4 each being a mostly new API without having to compromise for backward compatibility.
I think that's the price to pay for trying to cover a wide range of hardware. You can't just make all those shitty Android phones disappear. At least for each WebGPU limit, there's usually a Github ticket which explains why exactly this limit exists.
Both WebGL2 and WebGPU are probably the most 'watertight' specced and tested 3D API ever built, and especially WebGPU has gone to great lengths to eliminate UB present in native APIs (even at the cost of usability).
We only need to open chrome://gpu and see how many workarounds are implementated.
Those that happen to own a device where workarounds are yet to be implemented, have quite interesting experiences, depending on the root cause.
As this is an increasing list across Chrome releases.
Lets see how it works out there with Firefox and Safari, the later still not fully WebGL 2.0 compliant.
So much for the watertightness.
What I really would like to see is browser vendors finally providing WebGL and WebGPU debugging tools.
I think a decade has been more than enough for that.
Then again, no one is paying for browsers, so I guess I should not complain.
[0]: https://github.com/Traverse-Research/gpu-allocator/
[1]: https://www.evolvebenchmark.com/
As I also depend on the wgpu-native bindings it's slow for updates to reach. Like we just got to v25 last week and v26 dropped a couple days prior to that.
I just installed the Mac nightly from https://www.mozilla.org/en-US/firefox/channel/desktop/ and now this demo works: https://huggingface.co/spaces/reach-vb/github-issue-generato...
It runs the SmolLM2 model compiled to WebAssembly for structured data extraction. I previously thought that demo only worked in Chrome.
(If I try it in regular Firefox for Mac I get "Error: WebGPU is not supported in your current environment, but it is necessary to run the WebLLM engine.")
> Although Firefox 141 enables WebGPU only on Windows, we plan to ship WebGPU on Mac and Linux in the coming months, and finally on Android.
Sounds good. I'm not really thrilled about it as of now. What ever the reason, it's not been supported in Linux for any browsers as of yet. My guess is it's too hard to expose without creating terrible attack surfaces.
This seems to support my view that web standards are too overgrown for how users actually use the web. It's obviously too late to do anything about it now but all the issues of monoculture and funding we are worried about today stem from the complexity of making a web browser due to decisions tracing all the way back to the days of Netscape.
However it kind of proves the point on how relevant browser vendors see GNU/Linux for this kind of workloads.
Gaussian splatting training and rendering using webgpu
1. https://boat-demo.cds.unity3d.com/
2. https://www.keijiro.tokyo/WebGPU-Test/
3. https://www.chatlord.com/4/
[0] https://compute.toys/
[1] https://github.com/mikbry/awesome-webgpu
[2] https://github.com/s-macke/WebGPU-Lab
https://vester.si/motion/
It also works without WebGPU, just very slowly.
I was feeling a bit dirty playing around with WebGPU with only Chrome into the game thus far, even Safari has enabled their preview quite recently.
- Visualize other scan data such as gaussian splat data sets, or triangle meshes from photogrammetry
- Things like google earth, Cesium, or other 3D globe viewers.
It's a pretty big thing in geospatial sciences and industry.
For gaussian splatting, WebGPU is great since it allows implementing sorting via compute shaders. WebGL-based implementations sort on the CPU, which means "correct" front-to-back blending lags behind for a few frames.
But yeah, when you ask like that, it would have been much better if they had simply added compute shaders to WebGL, because other than that there really is no point in WebGPU.
While I would have designed a few things differently in WebGPU (especially around the binding model), it's still a much better API than WebGL2 from every angle.
The limited feature set of WebGPU is mostly to blame on Vulkan 1.0 drivers on Android devices I guess, but there's no realistic way to design a web 3D API and ignore shitty Android phones unfortunately.
The one thing that WebGPU is doing better is that it does implicit syncing by default. The problem is, it provides no options for explicit syncing.
I mainly software-rasterize everything in Cuda nowadays, which makes the complexity of graphics apis appear insane. Cuda allows you to get things done simple and easily, but it still has all the functionaility to make things fast and powerful. The important part is that the latter is optinal, so you can get things done quickly, and still make them fast.
In cuda, allocating a buffer and filling it with data is a simple cuMemAlloc and cuMemcpy. When calling a shader/kernel, I dont need bindings and descriptors, I simply pass a pointer to the data. Why would I need that anyway, the shader/kernel knows all about the data, the host doesnt need to know.
AFAIK Vulkan only eliminated pre-baked render pass objects (which were indeed pointless), and now simply copied Metal's design of transient render passes, e.g. there's still 'render pass boundaries' between vkCmdBeginRendering() and vkCmdEndRendering() and the VkRenderingInfo struct that's passed into the vkCmdBeginRendering() function (https://registry.khronos.org/vulkan/specs/latest/man/html/Vk...) is equivalent with Metal's MTLRenderPassDescriptor (https://developer.apple.com/documentation/metal/mtlrenderpas...).
E.g. even modern Vulkan still has render passes, they just didn't want to call those new functions 'Begin/EndRenderPass' for some reason ;) AFAIK the idea of render pass boundaries is quite essential for tiler GPUs.
WebGPU pretty much tries to copy Metal's render pass approach as much as possible (e.g. it doesn't have pre-baked pass objects like Vulkan 1.0).
> The one thing that WebGPU is doing better is that it does implicit syncing by default.
AFAIK also mostly thanks to the 'transient render pass model'.
> Why would I need that anyway, the shader/kernel knows all about the data, the host doesnt need to know.
Because old GPUs are a thing and those usually don't have such a flexible hardware design to make rasterizing (or even vertex pulling) in compute shaders performant enough to compete with the traditional render pipeline.
> Similarly static binding groups are entirely pointless
I agree, but AFAIK Vulkan's 1.0 descriptor model is mostly to blame for the inflexible BindGroups design.
> but that's also made needlessly cumbersome in WebGPU due to the requirement to use staging buffers
Most modern 3D APIs also switched to staging buffers though, and I guess there's not much choice if you don't have unified memory.
I've been told by a driver dev of a tiler GPU that they are, in fact, not essential. They pick that info up by themselves by analyzing the command buffer.
Well I wouldn't know since I switched to using Cuda as a graphics API. It's mostly nonsense-free, and faster than the hardware pipeline for points, and about as fast for splats. Seeing how Nanite also software-rasterizes as a performance improvement, Cuda may even be great for triangles. Only implemented a rudimentary triangle rasterizer that can draw 10 million small textured triangles per millisecond. Still working on the larger ones, but low-priority since I focus on point clouds.
In any case, I won't touch graphics APIs anymore until they make a clean break to remove the legacy nonsense. Allocating buffers should be a single line, providing data to shaders should be as simple as passing pointers, etc..
> Are we sure sites are not just going to use it to mine bitcoins using their users' hardware?
Some almost certainly will but like all similar issues the game of cat and mouse will continue.
https://en.wikipedia.org/wiki/Infinity_Blade
Game demo, https://www.youtube.com/watch?v=_w2CXudqc6c
The only thing I like in Web 3D APIs, is that outside middleware engines, they are the only mainstream 3D APIs designed with managed languages in mind, instead of after the fact bindings.
Still waiting for something like RenderDoc on the respective browser developer tools, we never got anything better than SpectorJS.
It isn't even printf debugging, rather pixel colour debugging.
And the SWF format had insane compatibility, literally unmatched by any other technology imo, we didn't even think about OS's, it really was "write once run anywhere" (pre-smartphone ofc). On the web, even basic CSS doesn't work the same from OS to OS, and WebGL apps still crash on 10% of devices randomly. It'll probably be 5 years before WebGPU is even remotely stable.
Not even to mention the fully integrated editor environment.
Or I guess maybe you're saying someone should build something like Flash targeting WebGPU? Probably the closest there is to that right now is Figma? But it feels weak too imo, and was already possible with WebGL. Maybe Unreal Engine is the bet.
Consequently much of the JS 3D community has become obsessed with gaussian splatting, and AR more generally.
[1] And I would extend this to what's going on here: people prefer complaining about how missing features in APIs prevent their genius idea from being possible, when in truth there's simply no demand from users for this stuff at all. You could absolutely have done web Minecraft years ago, and it's very revealing such a thing is not wildly popular. I personally wasted too long on WebGL ( https://www.luduxia.com/ ), and what I learned is the moment it all works people just assume it was nothing and move on.
Driver and OS blacklisting, means that game developers aren't aware of the user experience, nor can they controll it, as in native games, or server side rendering with streaming.
No proper debugging tools other than printf/pixel debugging.
The amount of loading screens that would be needed, given memory constraints of browser sessions.
This alone means there is hardly that much ROI for 3D webgames, and most uses end up being in ecommerce, or Google Maps kind of applications.
There is a tracking issue[1], although I ma not sure how much of that makes it to the browser.
[1] https://github.com/gfx-rs/wgpu/issues/6762
This might still be a semi-legitimate thing, i.e maybe they kept around a WebGL implementation for a while as a fallback but moved the main implementation to WebGPU and don't want to maintain the fallback. It certainly fits well into their strategy of making sure that the web really only works properly with Chrome.