Re:Claude Code Daily Benchmarks for Degradation Tracking
by bonoboTP on Thursday January 29, 2026 @05:29PM
(#46817717)
It can be made deterministic. It's not trivial and can slow it down a bit (not much) but there are environment variables you can set to make your GPU computations bitwise reproducible. I have done this in training models with Pytorch.
Re:Claude Code Daily Benchmarks for Degradation Tracking
by minimaltom on Thursday January 29, 2026 @06:47PM
(#46818589)
There are settings to make it reproducible but they incur a non-negligible drop in performance.
Unsurprising given they amount to explicit synchronization to make the order of operations deterministic.
When you add A then B then C, you get a different answer than C then A then B, because floating point, approximation error, subnormals etc.