ClimaLand.jl icon indicating copy to clipboard operation
ClimaLand.jl copied to clipboard

Variability in average time on benchmark runs

Open kmdeck opened this issue 8 months ago • 1 comments

When running the Richards and Bucket benchmark CI runs, the computed average time in a build can vary by large amounts (larger than the standard deviation within a build) from build to build.

Is this due to running on different GPUs? or is it an underlying issue?

kmdeck avatar Jun 16 '24 23:06 kmdeck