+- 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: when TensorRT-LLM clearly wins: early impressions (/showthread.php?tid=576)
Pages:
1
2
when TensorRT-LLM clearly wins: early impressions - vectorport - 09-29-2026
Thought I'd post a small recipe because a lot of threads ask similar things.
This isn't revolutionary, but the combination of SGLang and 5090 felt surprisingly usable once I tuned around workstation thermals.
Maybe useful as a starting point for newer folks.
This is mostly for local work rather than public serving.
Re: when TensorRT-LLM clearly wins: early impressions - rocmstack25 - 09-30-2026
A couple things I'd check first. Before rebuilding anything, I'd test one clean baseline with Qwen and compare from there.
Re: when TensorRT-LLM clearly wins: early impressions - sysstation69 - 09-30-2026
A couple things I'd check first. Start by verifying the driver/runtime combo, then test with a smaller prompt, then look at INT4. Once you know the plain path works, then add the faster path back in piece by piece.
Re: when TensorRT-LLM clearly wins: early impressions - turbosupply83 - 09-30-2026
Quote:A couple things I'd check first. Start by verifying the driver/runtime combo, then test with a smaller prompt, then look at INT4. Once you know the pl
Appreciate the detail. That matches what I saw on 5090. Worth rechecking your defaults though, because one bad assumption can make the rest of the thread look confusing.
Re: when TensorRT-LLM clearly wins: early impressions - pixelcrate82 - 09-30-2026
My first guess would be this: If you're on llama.cpp, I'd confirm the version first and then reduce variables until the behavior is obvious. A lot of these threads end up being one bad config plus a pile of conclusions drawn from it.
Re: when TensorRT-LLM clearly wins: early impressions - relayforge - 10-01-2026
A couple questions before I guess too hard. Do you see the same thing with a smaller batch or shorter prompt?
Re: when TensorRT-LLM clearly wins: early impressions - vectorport - 10-01-2026
Thanks, that helps. 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.
Re: when TensorRT-LLM clearly wins: early impressions - kernelmike - 10-01-2026
Need a little more detail here. Are you testing this on dual 3090 or something else?
Re: when TensorRT-LLM clearly wins: early impressions - sysstation69 - 10-01-2026
Thanks, that helps. That matches what I saw on Threadripper box. Worth rechecking your defaults though, because one bad assumption can make the rest of the thread look confusing.
Re: when TensorRT-LLM clearly wins: early impressions - kernelmike - 10-01-2026
Quote:Thanks, that helps. That matches what I saw on Threadripper box. Worth rechecking your defaults though, because one bad assumption can make the rest o
worth checking the driver first.