LLVM Clang 6.0 Benchmarks On AMD's EPYC Yield Some Performance Benefits

Comparing the different micro-architecture targets show "znver1" overall performing well. Using "-march=znver1" overall yielded the fastest binaries as it should when running on AMD EPYC/Ryzen processors. But in a few select test-cases, the Haswell target was yielding slightly faster performance or in very rare cases even the generic x86-64 target. You can dig through those numbers via this OpenBenchmarking.org result file.

Overall, LLVM Clang 6.0 has better support for AMD Zen (znver1) processors than last year's Clang 5.0 update and if you are compiling your own applications that are performance-sensitive, it is worth experimenting with the latest compiler releases when available. As it stands now LLVM Clang 6.0.0 should be officially released by the end of February.

For those wondering how the results of LLVM Clang 6.0 SVN compare to the in-development GCC 8.0, I am presently working on fresh benchmark results there to share in the days ahead.