+- Inference (https://www.datasculpting.ai/inference)
+-- Forum: Evaluation & Testing (https://www.datasculpting.ai/inference/forumdisplay.php?fid=16)
+--- Forum: Quality / Reliability Testing (https://www.datasculpting.ai/inference/forumdisplay.php?fid=68)
+--- Thread: when video is still not worth the wait (/showthread.php?tid=78)
when video is still not worth the wait - bytehall - 08-31-2026
Posting this in case it saves someone else a little time.
I'm not trying to build a perfect architecture diagram. I just want a sane starting point around llama.cpp and Threadripper box.
I'm especially interested in mistakes worth avoiding.
I care more about stability than squeezing the last possible tokens/sec out of it.
Re: when video is still not worth the wait - plainember52 - 08-31-2026
What worked for me was pretty simple. On A6000, 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.
Re: when video is still not worth the wait - fusedriver - 08-31-2026
I tested this tonight. I couldn't reproduce the slowdown until I pushed context much higher.
Re: when video is still not worth the wait - phaseport36 - 08-31-2026
I'd approach it in a boring order. If you're on vLLM, 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 video is still not worth the wait - sparrowshed - 08-31-2026
A couple things I'd check first. Before rebuilding anything, I'd test one clean baseline with Llama and compare from there.
Re: when video is still not worth the wait - labline - 09-03-2026
That matches what I was seeing. That matches what I saw on GB200. I got better results once I stopped chasing one metric and just looked at whether the workflow felt usable.
Re: when video is still not worth the wait - audiodesk - 09-06-2026
I don't think that's the whole story. The reported numbers can look compelling, but in ordinary usage I didn't see the same benefit. Not saying you're wrong, just that I wouldn't lock that in as the lesson yet.