研究社交
Maybe related to #3518 and #3498?
Update: Tried ```rust queue.submit(None); device.poll(Maintain::Wait); ``` before dropping, it gets mitigated, but each reload still adds 0.2GB.
Yes, this is the same problem.
Hi! I am currently developing [`web-rwkv`](https://github.com/cryscan/web-rwkv) which implements an LLM with WGPU. It's already fast, but having this feature could make it even faster, which is super nice.
`Ai00` requires Vulkan. Does your GPU support Vulkan and Do you have Vulkan driver installed? You may check `vulkaninfo`.
I don't know whether it's a limit imposed by your GPU. I got tested on some low-end NVidia GPUs and that was allowed. Could you try to run with `--release`?
If 16 is your limit then it should be resolved since v0.2.10
Thanks for informing. According to WGPU docs, [SAMPLED_TEXTURE_AND_STORAGE_BUFFER_ARRAY_NON_UNIFORM_INDEXING](https://docs.rs/wgpu/latest/wgpu/struct.Features.html#associatedconstant.SAMPLED_TEXTURE_AND_STORAGE_BUFFER_ARRAY_NON_UNIFORM_INDEXING) is supported on: - DX12 - Metal (with MSL 2.0+ on macOS 10.13+) - Vulkan 1.2+ I think your platform supports this...
Thank you for reminding!
The current ray tracing routine only considers geometry but not surface attributes. It is easy to render alpha-masked meshes correctly in the prepass (onto the G-Buffer) so their shape _appear_...