4090 vs 5090 for 70B quants with TensorRT-LLM

[turbogrid11]

I'm trying to sanity-check my setup.

My current setup is llama.cpp with 4090, and I keep running into workstation thermals.

Not looking for the perfect answer, mostly what worked in real use.

[vrambandit]

A couple things I'd check first. Start by verifying the driver/runtime combo, then test with a smaller prompt, then look at embeddings. A lot of these threads end up being one bad config plus a pile of conclusions drawn from it.

[rocmriley]

Tried to reproduce it just now. My result was mixed: first run looked bad, second run looked normal. The surprise was that memory pressure showed up before throughput collapsed. At this point the real bottleneck is my patience.

[cudapoint]

Thanks, that helps. I don't think your conclusion is far off. I got better results once I stopped chasing one metric and just looked at whether the workflow felt usable.

[vrambandit]

Appreciate the detail. 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. The logs were somehow calmer than I was.

[embersmith]

Tried to reproduce it just now. On GB200 with LM Studio, I saw basically the same behavior. The surprise was that memory pressure showed up before throughput collapsed.

[batchwire]

Not sure I agree there. The reported numbers can look compelling, but in ordinary usage I didn't see the same benefit.

[routecrate]

I tested this tonight. I couldn't reproduce the slowdown until I pushed context much higher.

[llmtrack]

Interesting, I hadn't tried that. I don't think your conclusion is far off. I got better results once I stopped chasing one metric and just looked at whether the workflow felt usable.

[amberhall]

That matches what I was seeing. The pattern around structured output has been pretty similar for me. I got better results once I stopped chasing one metric and just looked at whether the workflow felt usable.