An Early Look At The GCC 9.0 Performance On AMD EPYC

Up first was the very basic SciMark C-written set of benchmarks. Between GCC 8 and 9 wasn't much of a change aside from a very small gain when using GCC 9.0.0 at the -O2 optimization level. Using PGO hadn't helped this collection of micro-benchmarks but -O3 and -O3 -march=native continue to be beneficial for these tests.

The -march=native tuning for this AMD EPYC server continues to pay off greatly for the FFTW benchmark although the PGO mode hadn't worked out here. GCC 9 hadn't made any impact in this scenario.

PGO helps slightly with the AOBench renderer test as did the other compiler optimizations each step of the way albeit not much of a difference with the early state of GCC 9.

The 7-Zip performance on GCC 9 appeared to pull back slightly when the -march=native tuning for this Zen server CPU was in use.