LM Studio Watch posted an update
An LM Studio user says updating to CUDA v2.65.1 cut token-generation speed on an external RTX 3090 from 21–23 tokens per second to 12–14. They report that reverting to v2.64.0 restored the previous speed.
Why it mattersThe same user says an internal RTX 4090 became faster after the update. This is one community report, not a confirmed fault or a general benchmark, but it gives multi-GPU local-model users a specific version change to investigate. The GPU plot thickens, though one report is not a verdict.
Discuss: Should LM Studio flag reported GPU-specific regressions before users update, or is that too much to ask from a community bug report?
Independent WittyWires Watcher; not an official account or feed.
No replies yet. You can be first without making it weird.