GCC 7.3 vs. GCC 8.0 vs. LLVM Clang 6.0 On The POWER9 Raptor Talos II
As part of the remote testing of the Raptor Talos II Workstation that is comprised of fully free software down to the firmware and powered by high-end POWER9 processors, over Easter weekend I carried out some GCC vs. Clang benchmarks.
I'm still working on the POWER9 vs. Xeon vs. EPYC benchmarks on Debian Testing for publishing in the next day or two, but for those curious, below are some POWER9 compiler benchmark comparisons.
The stock compiler on Debian Testing was GCC 7.3.0 (though note it's --with-cpu= is configured by default for POWER8, not POWER9), then tested when building GCC 8.0.1 from source on 31 March though with configuring for the POWER9 processor explicitly, and then again when building the release source for LLVM Clang 6.0.
For the most part, LLVM Clang does a good job keeping up with GCC on 64-bit POWER9.
The Clang difference with CacheBench was certainly surprising and is to be investigated.
Clang 6.0 had problems building a few tests like 7-Zip on POWER9.
And in some computational tests, GCC still commands a sizable victory.
Between GCC 7.3 and GCC 8.0.1, the performance is largely the same even with switching to --with-cpu=power9 in the process.
In some workloads, GCC8 is faster on this free software system.
Clang was much faster with MP3 encoding using LAME on POWER.
GCC was slightly faster with OpenSSL signing.
Clang was slightly better with Redis.
Anyhow, while waiting for our Xeon vs. EPYC vs. POWER9 benchmark results, you can dig through more of these compiler benchmarks via this OpenBenchmarking.org result file.
I'm still working on the POWER9 vs. Xeon vs. EPYC benchmarks on Debian Testing for publishing in the next day or two, but for those curious, below are some POWER9 compiler benchmark comparisons.
The stock compiler on Debian Testing was GCC 7.3.0 (though note it's --with-cpu= is configured by default for POWER8, not POWER9), then tested when building GCC 8.0.1 from source on 31 March though with configuring for the POWER9 processor explicitly, and then again when building the release source for LLVM Clang 6.0.
For the most part, LLVM Clang does a good job keeping up with GCC on 64-bit POWER9.
The Clang difference with CacheBench was certainly surprising and is to be investigated.
Clang 6.0 had problems building a few tests like 7-Zip on POWER9.
And in some computational tests, GCC still commands a sizable victory.
Between GCC 7.3 and GCC 8.0.1, the performance is largely the same even with switching to --with-cpu=power9 in the process.
In some workloads, GCC8 is faster on this free software system.
Clang was much faster with MP3 encoding using LAME on POWER.
GCC was slightly faster with OpenSSL signing.
Clang was slightly better with Redis.
Anyhow, while waiting for our Xeon vs. EPYC vs. POWER9 benchmark results, you can dig through more of these compiler benchmarks via this OpenBenchmarking.org result file.
5 Comments