NVIDIA driver regression: early impressions

[sparrowkite53]

subscribed, curious where this lands.

I'd approach it in a boring order. If you're on MLX, I'd confirm the version first and then reduce variables until the behavior is obvious.

[kernelmike]

subscribed, curious where this lands.

Thanks, that helps. I don't think your conclusion is far off. Worth rechecking your defaults though, because one bad assumption can make the rest of the thread look confusing.

[cobaltspark]

subscribed, curious where this lands.

That matches what I was seeing. The pattern around NVFP4 has been pretty similar for me. The biggest change for me was simplifying the stack instead of adding one more optimization layer.

[audiopath66]

subscribed, curious where this lands.

Maybe, but I had the opposite result. I've seen people blame tool calling when the real issue was much more mundane. Worth testing the simpler explanation before adopting the more interesting one.

[dockember]

following this because I'm in the same boat.

[audiopath66]

Ran a quick check on my box. My result was mixed: first run looked bad, second run looked normal. The surprise was that memory pressure showed up before throughput collapsed.