SPEC CPU result comparison

15th October 2026 - Normalized report

SPEC2017 FPrate: Trunk native -Ofast - standard vs PGO vs LTO+PGO

-Ofast -march=native -mtune=native

AMD EPYC 8635P (zen5 based Sorano)

Run-times

Benchmark Normal PGO LTO+PGO PGO LTO+PGO
503.bwaves_r 100.00% 97.81% 97.81% 100.00% 100.00%
507.cactuBSSN_r 100.00% 96.12% 96.12% 100.00% 100.00%
508.namd_r 100.00% 97.99% 101.01% 100.00% 103.08%
510.parest_r 100.00% 100.83% 97.92% 100.00% 97.11%
511.povray_r 100.00% 93.53% 73.90% 100.00% 79.01%
519.lbm_r 100.00% 100.00% 100.00% 100.00% 100.00%
521.wrf_r 100.00% 96.97% 96.97% 100.00% 100.00%
526.blender_r 100.00% 102.13% 102.84% 100.00% 100.69%
527.cam4_r 100.00% 93.17% 91.26% 100.00% 97.95%
538.imagick_r 100.00% 157.71% 159.43% 100.00% 101.09%
544.nab_r 100.00% 96.67% 98.33% 100.00% 101.72%
549.fotonik3d_r 100.00% 100.44% 102.87% 100.00% 102.42%
554.roms_r 100.00% 100.00% 102.49% 100.00% 102.49%
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 Normal PGO LTO+PGO PGO LTO+PGO
503.bwaves_r 100.00% 91.87% 88.41% 100.00% 96.23%
507.cactuBSSN_r 100.00% 55.91% 47.97% 100.00% 85.81%
508.namd_r 100.00% 48.66% 42.32% 100.00% 86.96%
510.parest_r 100.00% 45.61% 11.32% 100.00% 24.83%
511.povray_r 100.00% 72.83% 68.93% 100.00% 94.64%
519.lbm_r 100.00% 93.91% 80.11% 100.00% 85.31%
521.wrf_r 100.00% 81.25% 62.65% 100.00% 77.11%
526.blender_r 100.00% 76.05% 69.71% 100.00% 91.67%
527.cam4_r 100.00% 97.58% 37.06% 100.00% 37.98%
538.imagick_r 100.00% 44.26% 29.34% 100.00% 66.29%
544.nab_r 100.00% 87.46% 44.05% 100.00% 50.37%
549.fotonik3d_r 100.00% 65.93% 60.21% 100.00% 91.33%
554.roms_r 100.00% 91.91% 56.19% 100.00% 61.14%
Smaller binary sizes are always better.

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

Run-times

Benchmark Normal PGO LTO+PGO PGO LTO+PGO
503.bwaves_r 100.00% 96.44% 95.02% 100.00% 98.52%
507.cactuBSSN_r 100.00% 98.62% 98.62% 100.00% 100.00%
508.namd_r 100.00% 100.70% 99.30% 100.00% 98.61%
510.parest_r 100.00% 101.16% 98.84% 100.00% 97.70%
511.povray_r 100.00% 88.77% 77.34% 100.00% 87.12%
519.lbm_r 100.00% 99.15% 99.15% 100.00% 100.00%
521.wrf_r 100.00% 97.95% 97.37% 100.00% 99.40%
526.blender_r 100.00% 97.51% 102.70% 100.00% 105.32%
527.cam4_r 100.00% 96.46% 92.93% 100.00% 96.34%
538.imagick_r 100.00% 133.18% 137.22% 100.00% 103.03%
544.nab_r 100.00% 93.14% 94.46% 100.00% 101.42%
549.fotonik3d_r 100.00% 98.13% 101.87% 100.00% 103.81%
554.roms_r 100.00% 100.00% 101.43% 100.00% 101.43%
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% 92.52% 86.34% 100.00% 93.32%
507.cactuBSSN_r 100.00% 61.65% 52.59% 100.00% 85.29%
508.namd_r 100.00% 49.15% 43.69% 100.00% 88.89%
510.parest_r 100.00% 48.41% 12.07% 100.00% 24.92%
511.povray_r 100.00% 73.81% 69.80% 100.00% 94.56%
519.lbm_r 100.00% 94.99% 82.34% 100.00% 86.69%
521.wrf_r 100.00% 80.44% 62.02% 100.00% 77.10%
526.blender_r 100.00% 76.73% 70.30% 100.00% 91.61%
527.cam4_r 100.00% 97.65% 37.10% 100.00% 37.99%
538.imagick_r 100.00% 48.59% 32.27% 100.00% 66.42%
544.nab_r 100.00% 88.06% 44.22% 100.00% 50.21%
549.fotonik3d_r 100.00% 64.47% 59.03% 100.00% 91.57%
554.roms_r 100.00% 91.19% 56.32% 100.00% 61.76%
Smaller binary sizes are always better.