SPEC CPU result comparison

15th October 2026 - Normalized report

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

-O2

AMD EPYC 8635P (zen5 based Sorano)

Run-times

Benchmark Normal LTO LTO+PGO LTO LTO+PGO
503.bwaves_r 100.00% 99.50% 98.76% 100.00% 99.25%
507.cactuBSSN_r 100.00% 101.77% 100.88% 100.00% 99.13%
508.namd_r 100.00% 100.00% 101.65% 100.00% 101.65%
510.parest_r 100.00% 96.83% 86.24% 100.00% 89.07%
511.povray_r 100.00% 93.72% 78.35% 100.00% 83.60%
519.lbm_r 100.00% 100.00% 100.00% 100.00% 100.00%
521.wrf_r 100.00% 100.00% 95.34% 100.00% 95.34%
526.blender_r 100.00% 97.99% 110.96% 100.00% 113.24%
527.cam4_r 100.00% 98.44% 90.42% 100.00% 91.86%
538.imagick_r 100.00% 94.78% 99.60% 100.00% 105.08%
544.nab_r 100.00% 100.00% 98.64% 100.00% 98.64%
549.fotonik3d_r 100.00% 96.91% 99.59% 100.00% 102.76%
554.roms_r 100.00% 100.80% 86.17% 100.00% 85.49%
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 LTO LTO+PGO LTO LTO+PGO
503.bwaves_r 100.00% 96.55% 106.21% 100.00% 110.00%
507.cactuBSSN_r 100.00% 86.95% 69.04% 100.00% 79.40%
508.namd_r 100.00% 99.92% 64.67% 100.00% 64.72%
510.parest_r 100.00% 20.90% 15.31% 100.00% 73.26%
511.povray_r 100.00% 100.63% 84.16% 100.00% 83.63%
519.lbm_r 100.00% 81.85% 91.65% 100.00% 111.97%
521.wrf_r 100.00% 76.29% 70.34% 100.00% 92.20%
526.blender_r 100.00% 98.67% 79.95% 100.00% 81.03%
527.cam4_r 100.00% 37.80% 37.63% 100.00% 99.55%
538.imagick_r 100.00% 69.70% 52.09% 100.00% 74.74%
544.nab_r 100.00% 49.94% 48.14% 100.00% 96.40%
549.fotonik3d_r 100.00% 96.67% 78.31% 100.00% 81.01%
554.roms_r 100.00% 67.85% 72.02% 100.00% 106.16%
Smaller binary sizes are always better.

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

Run-times

Benchmark LTO LTO+PGO
503.bwaves_r 100.00% 93.02%
507.cactuBSSN_r 100.00% 104.27%
508.namd_r 100.00% 101.23%
510.parest_r 100.00% 95.29%
511.povray_r 100.00% 87.63%
519.lbm_r 100.00% 101.27%
521.wrf_r 100.00% 94.15%
526.blender_r 100.00% 100.62%
527.cam4_r 100.00% 93.18%
538.imagick_r 100.00% 102.49%
544.nab_r 100.00% 97.85%
549.fotonik3d_r 100.00% 97.82%
554.roms_r 100.00% 84.72%
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 LTO LTO+PGO
503.bwaves_r 100.00% 109.86%
507.cactuBSSN_r 100.00% 79.40%
508.namd_r 100.00% 64.72%
510.parest_r 100.00% 73.26%
511.povray_r 100.00% 83.64%
519.lbm_r 100.00% 111.92%
521.wrf_r 100.00% 92.20%
526.blender_r 100.00% 81.03%
527.cam4_r 100.00% 99.55%
538.imagick_r 100.00% 74.74%
544.nab_r 100.00% 96.42%
549.fotonik3d_r 100.00% 81.01%
554.roms_r 100.00% 106.16%
Smaller binary sizes are always better.