video model setup pain points on GB200

[thetashed24]

Did a bit of testing and thought I'd post the result here.

After a few runs on GB200, I ended up with decent performance, but the setup only really made sense once I adjusted around workstation thermals.

Curious if anyone else landed in the same place.

[mintforge]

My first guess would be this: Start by verifying the driver/runtime combo, then test with a smaller prompt, then look at INT4.

[vrambandit]

My first guess would be this: Start by verifying the driver/runtime combo, then test with a smaller prompt, then look at INT4.

What worked for me was pretty simple. If you're on llama.cpp, I'd confirm the version first and then reduce variables until the behavior is obvious. Once you know the plain path works, then add the faster path back in piece by piece.

[southtinker]

I'd approach it in a boring order. On 5090, the easiest win is often to back off the aggressive settings and see whether the problem survives. A lot of these threads end up being one bad config plus a pile of conclusions drawn from it.

[kernelgrid]

I don't think that's the whole story. I wouldn't optimize around that first. On my side the bigger swing came from the runtime and not the model. Worth testing the simpler explanation before adopting the more interesting one.

[labline]

I don't think that's the whole story. I get why that seems right, but on B200 the bottleneck felt elsewhere. Not saying you're wrong, just that I wouldn't lock that in as the lesson yet.

[rasterbench]

My first guess would be this: If you're on PyTorch, I'd confirm the version first and then reduce variables until the behavior is obvious.

[batchwire]

My first guess would be this: Start by verifying the driver/runtime combo, then test with a smaller prompt, then look at prefix caching.

[bytelab]

My first guess would be this: Before rebuilding anything, I'd test one clean baseline with Kimi and compare from there. Not glamorous advice, but repeatable baselines save a lot of time.

[southtinker]

I think the bottleneck is somewhere else. The reported numbers can look compelling, but in ordinary usage I didn't see the same benefit. Not saying you're wrong, just that I wouldn't lock that in as the lesson yet.