PCIe lane weirdness with inference loads: early impressions

[torchshed]

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.

[turbogrid11]

Good point. I ended up in about the same place using MLX.

[turboyard94]

subscribed, curious where this lands.

[turboarch95]

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.

[dockkite69]

Can you fill in a bit more context? Is this local-only or are you routing anything externally?

[vramarch]

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.

[signaltrack44]

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.