Threadripper build for inference: worth it?

[phasenode]

Spent some time on this today and wanted to compare notes.

Two paths both seem viable, but they break down differently once HBM constraints becomes important.

What would you pick if you had to live with it for a few months?

[vrampilot]

Did a small comparison this afternoon. I couldn't reproduce the slowdown until I pushed context much higher. The surprise was that memory pressure showed up before throughput collapsed.

[vramscope]

Ran a quick check on my box. On A6000 with LM Studio, I saw basically the same behavior.

[phasenode]

A couple things I'd check first. Start by verifying the driver/runtime combo, then test with a smaller prompt, then look at quant drift.

[vramghost]

Appreciate the detail. I ended up in about the same place using vLLM. Worth rechecking your defaults though, because one bad assumption can make the rest of the thread look confusing.

[vectorcrate98]

What worked for me was pretty simple. If you're on vLLM, I'd confirm the version first and then reduce variables until the behavior is obvious.

[labnest]

What worked for me was pretty simple. If you're on vLLM, I'd confirm the version first and then reduce variables until the behavior is obvious.

I'd approach it in a boring order. On dual 3090, 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.

There are a few different conversations hiding inside this thread, so being explicit about the goal helps.

[edgequill]

My first guess would be this: On GB200, the easiest win is often to back off the aggressive settings and see whether the problem survives. Once you know the plain path works, then add the faster path back in piece by piece.

[kernelhouse70]

I think the bottleneck is somewhere else. I've seen people blame tool calling when the real issue was much more mundane. Could be a workload difference, but I wouldn't generalize too fast from one setup.