mwyrzykowski

> overhead of re-recording the dispatches every frame Do you mean compute dispatches like dispatchWorkgroups @fridenmf ? Would it help if we had compute bundles too as mentioned above?

A tiny repro illustrating the overhead would help immensely to justify this too, if it can be provided.

> As for the RenderBundles, I don't know the whole story, but it seems like it's meant to model something like secondary command buffers in Vulkan which can be re-executed...

> @mwyrzykowski Just a heads-up: there’s currently an issue in the official Three.js BatchedMesh WebGPU example where setting the count above 1024 causes a break in the WebGPU backend of...

I did not try on D3D. I suppose as an optional extension, it could be unsupported on those backends. Alternatively, it is possible to emulate this functionality, but the implementation...

> That said, built-in printf functionality like @mwyrzykowski proposed would still be useful to have some batteries included, I just don't know where we would fit the additional print buffer...

I am unable to attend the working group meeting today, but either M2 or M3+ would be perfectly reasonable.

This might be a duplicate of https://github.com/gpuweb/gpuweb/issues/4348

Given the specification isn't at Draft CR stage yet, I don't understand the motivation for not making the potentially very minor breaking change here. If an implementation really wanted to...

WebLLM lowered the buffer size to 256MB. iOS is too likely to jetsam at 1GB - https://github.com/mlc-ai/web-llm/issues/386#issuecomment-2232281597 No change needed on the WebKit side