Any lessons to be learned from EasyPerf?

[cwpearson]

https://easyperf.net/blog/2019/08/02/Perf-measurement-environment-on-Linux

[abduld]

You might also want to look at https://github.com/rai-project/tensorflow/blob/master/tensorflow-agent/scripts/benchmarkModeEnable.sh since it has some effects on benchmarking. It's probably more pronounced for one-off testing rather than running the benchmark a few hundred times.