GCC 10 PGO Benchmarks On AMD Ryzen Threadripper 3960X + Ubuntu 19.10
For those looking for some fresh reference numbers on the impact of using GCC's Profile Guided Optimizations (PGO), here are some benchmark runs looking at the GCC 10 PGO performance on an Ubuntu 19.10 workstation built around the Ryzen Threadripper 3960X.
I also have some fresh GCC 10 LTO optimization benchmarks coming in the next few days but using the PGO PTS module I ran some fresh benchmarks with just "-O3 -march=native" and then again after making use of profile guided optimizations for the benchmark runs.
More GCC 10 benchmarks will be coming as the stable release approaches towards the end of Q1'2020.
Of the rather random assortment of tests run, TSCP saw the biggest help from enabling PGO in 13% better performance for this chess benchmark.
The MT-DGEMM benchmark was also close to 13% faster after being PGO'ed.
The PGO impact for various other tests ran ranged from no measurable change up to just a couple percent better performance. More tests beyond these were ran but they ended up being even less interesting. Stay tuned for more end-of-year GCC and Clang benchmarks soon. If you enjoy our different open-source/Linux benchmarks you can't find readily elsewhere, consider showing your support this holiday season.
I also have some fresh GCC 10 LTO optimization benchmarks coming in the next few days but using the PGO PTS module I ran some fresh benchmarks with just "-O3 -march=native" and then again after making use of profile guided optimizations for the benchmark runs.
More GCC 10 benchmarks will be coming as the stable release approaches towards the end of Q1'2020.
Of the rather random assortment of tests run, TSCP saw the biggest help from enabling PGO in 13% better performance for this chess benchmark.
The MT-DGEMM benchmark was also close to 13% faster after being PGO'ed.
The PGO impact for various other tests ran ranged from no measurable change up to just a couple percent better performance. More tests beyond these were ran but they ended up being even less interesting. Stay tuned for more end-of-year GCC and Clang benchmarks soon. If you enjoy our different open-source/Linux benchmarks you can't find readily elsewhere, consider showing your support this holiday season.
14 Comments