+- Inference (https://www.datasculpting.ai/inference)
+-- Forum: Hardware (https://www.datasculpting.ai/inference/forumdisplay.php?fid=4)
+--- Forum: GPUs (https://www.datasculpting.ai/inference/forumdisplay.php?fid=28)
+--- Thread: PCIe lane weirdness with inference loads: early impressions (/showthread.php?tid=855)
PCIe lane weirdness with inference loads: early impressions - torchshed - 10-07-2026
Thought I'd post a small recipe because a lot of threads ask similar things.
Just sharing a setup that ended up working reasonably well: Mixtral, MLX, dual 3090, with a lot of attention on structured output.
Maybe useful as a starting point for newer folks.
For context, I'm not chasing leaderboard numbers. I mostly want the system to feel consistent day to day.
Re: PCIe lane weirdness with inference loads: early impressions - turbogrid11 - 10-07-2026
Good point. I ended up in about the same place using MLX.
Re: PCIe lane weirdness with inference loads: early impressions - turboyard94 - 10-07-2026
subscribed, curious where this lands.
Re: PCIe lane weirdness with inference loads: early impressions - turboarch95 - 10-08-2026
Tried to reproduce it just now. My result was mixed: first run looked bad, second run looked normal. The main thing that moved was latency, not quality.
Re: PCIe lane weirdness with inference loads: early impressions - dockkite69 - 10-08-2026
Can you fill in a bit more context? Is this local-only or are you routing anything externally?
Re: PCIe lane weirdness with inference loads: early impressions - vramarch - 10-08-2026
That matches what I was seeing. That matches what I saw on B200. Worth rechecking your defaults though, because one bad assumption can make the rest of the thread look confusing.
Re: PCIe lane weirdness with inference loads: early impressions - signaltrack44 - 10-08-2026
A couple things I'd check first. Before rebuilding anything, I'd test one clean baseline with Kimi and compare from there. A lot of these threads end up being one bad config plus a pile of conclusions drawn from it.