Asked about Knowing which run produced which result
Two runs with the same seed give different results. Why?
An answer that works
GPU operations are not always deterministic: parallel reductions can vary in order and some algorithms are selected at runtime. You can force determinism at a speed cost, but usually it is better to accept the variance and report a mean across seeds.
Reading an answer is not being able to give one. The section that teaches this is Knowing which run produced which result, 4 hours of Experiment tracking.
Learn this properly