5 ms·
This minimalism is very effective. I took the opposite approach, and it has cause great pain. I've been writing a metaverse client in Rust. Right now, it's run
by Animats 2y ago
This minimalism is very effective.
I took the opposite approach, and it has cause great pain. I've been writing a metaverse client in Rust. Right now, it's running on another screen, showing an avatar riding a tram through a large steampunk city. I let that run for 12 hours before shipping a new pre-release.
This uses Vulkan, but it has WGPU and Rend3 on top. Rend3 offers a very clean API - you create meshes, 2d textures, etc., and "objects", which reference the meshes and textures. Creating an object puts it on screen. Rust reference counting interlocks everything. It's very straightforward to use.
All those layers create problems. WGPU tries to support web browsers, Vulkan, Metal, DX11 (recently dropped), DX12, Android, and OpenGL. So it needs a big dev team and changes are hard. WGPU's own API is mostly like Vulkan - you still have to do your own GPU memory allocation and synchronization.
WGPU has lowest-common-denominator problems. Some of those platforms can't support some functions. WGPU doesn't support multiple threads updating GPU memory without interference, which Vulkan supports. That's how you get content into the GPU without killing the frame rate. Big-world games and clients need that. Also, having to deal with platforms with different concurrency restrictions results in lock conflicts that can kill performance.
Rend3 is supposed to be a modest level of glue code to handle synchronization and allocation. Those are hard to do in a general way. Especially synchronization. Rend3 also does frustum culling (which is a big performance win; you're not rendering what's behind you) and tried to do occlusion culling (which was a performance lose because the compute to do that slowed things down). It also does translucency, which means a depth sort. (Translucent objects are a huge pain. I really need them; I work on worlds with lots of windows, which you can see out of and see in.)
The Rust 3D stack people are annoyed with me because I've been pounding on them to fix their stack for three years now. That's all volunteer. Vulkan has money behind it and enough users to keep it maintained. Rend3 was recently abandoned by its creator, so now I have to go inside that and fix it. Few people do anything elaborate on WGPU - mostly it's 2D games you could have done in Flash, or simple static 3D scenes. Commercial projects continue to use Unity or UE5.
If I went directly to Vulkan, I'd still have to write synchronization, allocation, frustrum culling, and translucency. So that's a big switch.
Incidentally, Vulkano, the wrapper over Vulkan and Metal, has lowest-common-denominator problems too. It doesn't allow concurrent updating of assets in the GPU. Both Vulkan and Metal support that. But, of course, Apple does it differently.
- dc443 2y ago> WGPU doesn't support multiple threads updating GPU memory without interference, which Vulkan supports. This is really helpful for me to learn about, this is a key thing I want to be able to get right for having a good experience. I really hope WGPU can find a way to add something for this as an extension.
- dc443 2y agoDo you know if these things I found offer any hope for being able to continue rendering a scene smoothly while we handle GPU memory management operations on worker threads? https://gfx-rs.github.io/2023/11/24/arcanization.html https://gfx-rs.github.io/2023/11/24/arcanization.html https://github.com/gfx-rs/wgpu/issues/5322 https://github.com/gfx-rs/wgpu/issues/5322
- jms55 2y agoThe actual issue is not CPU-side. The issue is GPU-side. The CPU feeds commands (CommandBuffers) telling the GPU what to do over a Queue. WebGPU/wgpu/dawn only have a single general purpose queue. Meaning any data upload commands (copyBufferToBuffer) you send on the queue block rendering commands from starting. The solution is multiple queues. Modern GPUs have a dedicated transfer/copy queue separate from the main general purpose queue. WebGPU/wgpu/dawn would need to add support for additional queues: https://github.com/gpuweb/gpuweb/issues?q=is%3Aopen+is%3Aissue+label%3Amulti-queue https://github.com/gpuweb/gpuweb/issues?q=is%3Aopen+is%3Aiss... There's also ReBAR/SMA, and unified memory (UMA) platforms to consider, but that gets even more complex.