+- Inference (https://www.datasculpting.ai/inference)
+-- Forum: Inference Engines (https://www.datasculpting.ai/inference/forumdisplay.php?fid=9)
+--- Forum: TensorRT-LLM (https://www.datasculpting.ai/inference/forumdisplay.php?fid=37)
+--- Thread: serving on Blackwell (/showthread.php?tid=414)
Pages:
1
2
serving on Blackwell - northcraft90 - 09-23-2026
Thought I'd post a small recipe because a lot of threads ask similar things.
Maybe useful as a starting point: vLLM on Threadripper box, keeping HBM constraints in mind from the start.
If someone wants the exact settings I can clean them up and post them.
This is mostly for local work rather than public serving.
Re: serving on Blackwell - batchcraft - 09-23-2026
following this because I'm in the same boat.
Re: serving on Blackwell - torchtrack24 - 09-23-2026
I'd start with the simpler stack.
Re: serving on Blackwell - batchcraft - 09-24-2026
I think the bottleneck is somewhere else. I wouldn't optimize around that first. On my side the bigger swing came from the runtime and not the model. Could be a workload difference, but I wouldn't generalize too fast from one setup.
Re: serving on Blackwell - routecrate - 09-24-2026
That matches what I was seeing. I don't think your conclusion is far off.
Re: serving on Blackwell - tensorriver - 09-24-2026
What worked for me was pretty simple. On mini PC, the easiest win is often to back off the aggressive settings and see whether the problem survives.
Re: serving on Blackwell - northcraft90 - 09-25-2026
What exact setup are you running? Is this local-only or are you routing anything externally?
Re: serving on Blackwell - kernelmike - 09-25-2026
following this because I'm in the same boat.
Re: serving on Blackwell - quiettoken - 09-26-2026
Thanks, that helps. That matches what I saw on A6000. The biggest change for me was simplifying the stack instead of adding one more optimization layer.
Re: serving on Blackwell - kernelmike - 09-26-2026
Quote:Thanks, that helps. That matches what I saw on A6000. The biggest change for me was simplifying the stack instead of adding one more optimization laye
Can you fill in a bit more context? Do you see the same thing with a smaller batch or shorter prompt?