throughput per watt reality check for Qwen

[deltapoint17]

Spent some time on this today and wanted to compare notes.

I've gone back and forth between vLLM and llama.cpp for Llama. The interesting difference for me has been token cost.

Trying to avoid rebuilding the whole thing twice.

[rasterbench]

Interesting, I hadn't tried that. The pattern around small local stacks has been pretty similar for me.

[fastbench]

A couple things I'd check first. On GB200, 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.

[mlxgarden]

My first guess would be this: On 5090, the easiest win is often to back off the aggressive settings and see whether the problem survives.

[vrambandit]

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. I keep promising myself this is the last rebuild.

[mangohouse]

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 t

Appreciate the detail. I don't think your conclusion is far off. The biggest change for me was simplifying the stack instead of adding one more optimization layer.

[vramridge]

Ran a quick check on my box. I couldn't reproduce the slowdown until I pushed context much higher. At this point the real bottleneck is my patience.

[tensorkite]

What exact setup are you running? Do you see the same thing with a smaller batch or shorter prompt?

[kernelmike]

that fixed it for me too.

[macrolab]

A couple things I'd check first. If you're on vLLM, I'd confirm the version first and then reduce variables until the behavior is obvious. Not glamorous advice, but repeatable baselines save a lot of time.