when TensorRT-LLM clearly wins on 5090

[turboscope]

Posting this in case it saves someone else a little time.

I'm helping a small team set up a local workflow. We don't need a giant stack, but we do care about long context.

Would rather learn from your scars than collect my own all at once.

I care more about stability than squeezing the last possible tokens/sec out of it.

[turboscope5]

Appreciate the detail. 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.

[sysstation69]

Appreciate the detail. I ended up in about the same place using vLLM. Worth rechecking your defaults though, because one bad assumption can make the rest of the thread look confusing.

There are a few different conversations hiding inside this thread, so being explicit about the goal helps.

[southden]

Can you fill in a bit more context? What model, what runtime, and roughly what context size?

[paperrunner]

I hit almost the same issue last night. The logs were somehow calmer than I was.

[rasterbench]

Did a small comparison this afternoon. I couldn't reproduce the slowdown until I pushed context much higher.

[pixelcrate82]

Not sure I agree there. I've seen people blame KV cache pressure when the real issue was much more mundane. Could be a workload difference, but I wouldn't generalize too fast from one setup.

[batchcraft]

My first guess would be this: On Threadripper box, the easiest win is often to back off the aggressive settings and see whether the problem survives. Once you know the plain path works, then add the faster path back in piece by piece.

[nanopatch]

What worked for me was pretty simple. If you're on vLLM, I'd confirm the version first and then reduce variables until the behavior is obvious.

[batchcraft]

worth checking the driver first.