![]() |
|
PCIe lane weirdness with inference loads for Phi - Printable Version +- 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 for Phi (/showthread.php?tid=157) |
PCIe lane weirdness with inference loads for Phi - amberhall - 09-07-2026 This keeps coming up in conversation so I figured I'd start a thread. The more I work with local setups, the more it seems like structured output decides everything downstream. Curious whether that lines up with what other people are seeing. 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 for Phi - amberhall - 09-07-2026 I think the bottleneck is somewhere else. I get why that seems right, but on 5090 the bottleneck felt elsewhere. Could be a workload difference, but I wouldn't generalize too fast from one setup. Re: PCIe lane weirdness with inference loads for Phi - dockkite69 - 09-07-2026 I hit almost the same issue last night. Re: PCIe lane weirdness with inference loads for Phi - fusedhall - 09-08-2026 What exact setup are you running? Are you testing this on 7900 XTX or something else? Re: PCIe lane weirdness with inference loads for Phi - vectorpoint - 09-09-2026 Tried to reproduce it just now. The issue went away once I changed one setting and restarted the stack cleanly. Re: PCIe lane weirdness with inference loads for Phi - rocmriley - 09-09-2026 Interesting, I hadn't tried that. That matches what I saw on 4090. The biggest change for me was simplifying the stack instead of adding one more optimization layer. Re: PCIe lane weirdness with inference loads for Phi - audiorunner74 - 09-09-2026 What worked for me was pretty simple. Start by verifying the driver/runtime combo, then test with a smaller prompt, then look at small local stacks. Once you know the plain path works, then add the faster path back in piece by piece. If you post your exact settings, people here can probably narrow it down pretty fast. |