Running Some Fresh GCC 8.0 Compiler Benchmarks On AMD EPYC With "znver1"
As SUSE has been working in conjunction with AMD on more tuning for AMD Zen CPUs under the GCC compiler, here are some fresh benchmarks of the GCC 8 compiler code being tested on an AMD EPYC system.
For looking at the "znver1" improvements and general work that's ongoing for GCC 8 as the next annual GNU Compiler Collection update due out in 2018, I ran some fresh benchmarks of this GCC 8 compiler code after building it from source yesterday using its latest SVN/Git tip. The GCC 7.2 compiler was then built from source and configured in the same manner for showing how GCC 8.0 is currently comparing to GCC 7.2.
Ubuntu was running on this TYAN server with AMD EPYC 7601 during testing. During the Phoronix Test Suite testing process, the CFLAGS/CXXFLAGS were set to "-O3 -march=znver1" when testing both GCC 7.2 and GCC 8.0 SVN.
FFTW is running slightly faster with GCC 8.
SciMark's Monte Carlo is significantly faster now under GCC 8. Additional follow-up benchmarks will look more closely if it's due to znver1 optimizations or general improvements to the compiler code in GCC 8.
In most tests, the gains are quite small but measurable in a number of scenarios.
C-Ray seems to frequently become faster as a result of GCC optimizations.
LAME MP3 encoding is slightly faster according to the Phoronix Test Suite.
As outlined via this result file there were also some cases where GCC 8 isn't currently any different than GCC 7. More compiler benchmarks forthcoming on different AMD Zen processors as well as other compiler tuning tests in general as well as similar treatment under LLVM Clang.
For looking at the "znver1" improvements and general work that's ongoing for GCC 8 as the next annual GNU Compiler Collection update due out in 2018, I ran some fresh benchmarks of this GCC 8 compiler code after building it from source yesterday using its latest SVN/Git tip. The GCC 7.2 compiler was then built from source and configured in the same manner for showing how GCC 8.0 is currently comparing to GCC 7.2.
Ubuntu was running on this TYAN server with AMD EPYC 7601 during testing. During the Phoronix Test Suite testing process, the CFLAGS/CXXFLAGS were set to "-O3 -march=znver1" when testing both GCC 7.2 and GCC 8.0 SVN.
FFTW is running slightly faster with GCC 8.
SciMark's Monte Carlo is significantly faster now under GCC 8. Additional follow-up benchmarks will look more closely if it's due to znver1 optimizations or general improvements to the compiler code in GCC 8.
In most tests, the gains are quite small but measurable in a number of scenarios.
C-Ray seems to frequently become faster as a result of GCC optimizations.
LAME MP3 encoding is slightly faster according to the Phoronix Test Suite.
As outlined via this result file there were also some cases where GCC 8 isn't currently any different than GCC 7. More compiler benchmarks forthcoming on different AMD Zen processors as well as other compiler tuning tests in general as well as similar treatment under LLVM Clang.
9 Comments