SPEC CPU result comparison

15th October 2026 - Normalized report

SPEC2017 FPrate: Trunk generic -O2 - standard vs PGO vs LTO+PGO

-O2

AMD EPYC 8635P (zen5 based Sorano)

Run-times

Benchmark Normal PGO LTO+PGO PGO LTO+PGO
503.bwaves_r 100.00% 99.50% 98.76% 100.00% 99.25%
507.cactuBSSN_r 100.00% 99.41% 100.88% 100.00% 101.48%
508.namd_r 100.00% 100.82% 101.65% 100.00% 100.82%
510.parest_r 100.00% 88.36% 86.24% 100.00% 97.60%
511.povray_r 100.00% 96.32% 78.35% 100.00% 81.35%
519.lbm_r 100.00% 100.00% 100.00% 100.00% 100.00%
521.wrf_r 100.00% 95.34% 95.34% 100.00% 100.00%
526.blender_r 100.00% 104.03% 110.96% 100.00% 106.67%
527.cam4_r 100.00% 91.98% 90.42% 100.00% 98.31%
538.imagick_r 100.00% 100.00% 99.60% 100.00% 99.60%
544.nab_r 100.00% 98.64% 98.64% 100.00% 100.00%
549.fotonik3d_r 100.00% 96.91% 99.59% 100.00% 102.76%
554.roms_r 100.00% 83.78% 86.17% 100.00% 102.86%
The compiler used was GCC 17.0.0 20261015 (experimental) (revision 84348add236100e5f69a3e9ec4f7b1038bf2b83d).
Naturally, the smaller the run-times, the better.
See also continuous gcc options benchmarking at https://lnt.opensuse.org/db_default/v4/SPEC/spec_report/options.

Non-debug executable sizes

Benchmark Normal PGO LTO+PGO PGO LTO+PGO
503.bwaves_r 100.00% 113.16% 106.21% 100.00% 93.86%
507.cactuBSSN_r 100.00% 81.19% 69.04% 100.00% 85.03%
508.namd_r 100.00% 68.15% 64.67% 100.00% 94.89%
510.parest_r 100.00% 63.33% 15.31% 100.00% 24.18%
511.povray_r 100.00% 88.88% 84.16% 100.00% 94.70%
519.lbm_r 100.00% 107.02% 91.65% 100.00% 85.64%
521.wrf_r 100.00% 92.55% 70.34% 100.00% 76.00%
526.blender_r 100.00% 88.03% 79.95% 100.00% 90.83%
527.cam4_r 100.00% 99.11% 37.63% 100.00% 37.97%
538.imagick_r 100.00% 78.19% 52.09% 100.00% 66.62%
544.nab_r 100.00% 96.24% 48.14% 100.00% 50.03%
549.fotonik3d_r 100.00% 84.85% 78.31% 100.00% 92.29%
554.roms_r 100.00% 116.32% 72.02% 100.00% 61.92%
Smaller binary sizes are always better.

AMD EPYC 75F3 (zen3 based high-frequency Milan)

Run-times

Benchmark PGO LTO+PGO
503.bwaves_r 100.00% 84.58%
507.cactuBSSN_r 100.00% 100.69%
508.namd_r 100.00% 100.00%
510.parest_r 100.00% 98.30%
511.povray_r 100.00% 91.67%
519.lbm_r 100.00% 100.63%
521.wrf_r 100.00% 98.94%
526.blender_r 100.00% 100.00%
527.cam4_r 100.00% 98.90%
538.imagick_r 100.00% 99.40%
544.nab_r 100.00% 100.00%
549.fotonik3d_r 100.00% 100.90%
554.roms_r 100.00% 99.49%
The compiler used was GCC 17.0.0 20261015 (experimental) (revision 84348add236100e5f69a3e9ec4f7b1038bf2b83d).
Naturally, the smaller the run-times, the better.
See also continuous gcc options benchmarking at https://lnt.opensuse.org/db_default/v4/SPEC/spec_report/options.

Non-debug executable sizes

Benchmark PGO LTO+PGO
503.bwaves_r 100.00% 93.81%
507.cactuBSSN_r 100.00% 85.03%
508.namd_r 100.00% 94.89%
510.parest_r 100.00% 24.18%
511.povray_r 100.00% 94.70%
519.lbm_r 100.00% 85.69%
521.wrf_r 100.00% 76.00%
526.blender_r 100.00% 90.83%
527.cam4_r 100.00% 37.97%
538.imagick_r 100.00% 66.62%
544.nab_r 100.00% 50.03%
549.fotonik3d_r 100.00% 92.30%
554.roms_r 100.00% 61.92%
Smaller binary sizes are always better.