SPEC CPU result comparison

15th October 2026 - Normalized report

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

-Ofast -march=native -mtune=native

AMD EPYC 8635P (zen5 based Sorano)

Run-times

Benchmark Normal LTO LTO+PGO LTO LTO+PGO
503.bwaves_r 100.00% 97.81% 97.81% 100.00% 100.00%
507.cactuBSSN_r 100.00% 100.00% 96.12% 100.00% 96.12%
508.namd_r 100.00% 100.00% 101.01% 100.00% 101.01%
510.parest_r 100.00% 96.25% 97.92% 100.00% 101.73%
511.povray_r 100.00% 92.38% 73.90% 100.00% 80.00%
519.lbm_r 100.00% 101.42% 100.00% 100.00% 98.60%
521.wrf_r 100.00% 100.00% 96.97% 100.00% 96.97%
526.blender_r 100.00% 99.29% 102.84% 100.00% 103.57%
527.cam4_r 100.00% 98.09% 91.26% 100.00% 93.04%
538.imagick_r 100.00% 93.14% 159.43% 100.00% 171.17%
544.nab_r 100.00% 100.00% 98.33% 100.00% 98.33%
549.fotonik3d_r 100.00% 100.00% 102.87% 100.00% 102.87%
554.roms_r 100.00% 99.00% 102.49% 100.00% 103.52%
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% 101.01% 88.41% 100.00% 87.53%
507.cactuBSSN_r 100.00% 92.37% 47.97% 100.00% 51.94%
508.namd_r 100.00% 97.99% 42.32% 100.00% 43.18%
510.parest_r 100.00% 18.71% 11.32% 100.00% 60.53%
511.povray_r 100.00% 106.07% 68.93% 100.00% 64.99%
519.lbm_r 100.00% 75.71% 80.11% 100.00% 105.82%
521.wrf_r 100.00% 77.14% 62.65% 100.00% 81.22%
526.blender_r 100.00% 107.44% 69.71% 100.00% 64.89%
527.cam4_r 100.00% 38.12% 37.06% 100.00% 97.22%
538.imagick_r 100.00% 66.02% 29.34% 100.00% 44.44%
544.nab_r 100.00% 49.13% 44.05% 100.00% 89.65%
549.fotonik3d_r 100.00% 97.66% 60.21% 100.00% 61.65%
554.roms_r 100.00% 73.88% 56.19% 100.00% 76.06%
Smaller binary sizes are always better.

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

Run-times

Benchmark Normal LTO LTO+PGO LTO LTO+PGO
503.bwaves_r 100.00% 94.31% 95.02% 100.00% 100.75%
507.cactuBSSN_r 100.00% 101.10% 98.62% 100.00% 97.55%
508.namd_r 100.00% 100.00% 99.30% 100.00% 99.30%
510.parest_r 100.00% 96.90% 98.84% 100.00% 102.00%
511.povray_r 100.00% 91.89% 77.34% 100.00% 84.16%
519.lbm_r 100.00% 105.38% 99.15% 100.00% 94.09%
521.wrf_r 100.00% 99.42% 97.37% 100.00% 97.94%
526.blender_r 100.00% 102.70% 102.70% 100.00% 100.00%
527.cam4_r 100.00% 95.71% 92.93% 100.00% 97.10%
538.imagick_r 100.00% 96.86% 137.22% 100.00% 141.67%
544.nab_r 100.00% 98.68% 94.46% 100.00% 95.72%
549.fotonik3d_r 100.00% 100.47% 101.87% 100.00% 101.40%
554.roms_r 100.00% 99.29% 101.43% 100.00% 102.16%
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% 101.72% 86.34% 100.00% 84.88%
507.cactuBSSN_r 100.00% 93.11% 52.59% 100.00% 56.48%
508.namd_r 100.00% 98.04% 43.69% 100.00% 44.56%
510.parest_r 100.00% 19.45% 12.07% 100.00% 62.05%
511.povray_r 100.00% 105.97% 69.80% 100.00% 65.87%
519.lbm_r 100.00% 76.90% 82.34% 100.00% 107.08%
521.wrf_r 100.00% 76.97% 62.02% 100.00% 80.58%
526.blender_r 100.00% 106.92% 70.30% 100.00% 65.75%
527.cam4_r 100.00% 38.09% 37.10% 100.00% 97.40%
538.imagick_r 100.00% 66.49% 32.27% 100.00% 48.53%
544.nab_r 100.00% 48.92% 44.22% 100.00% 90.39%
549.fotonik3d_r 100.00% 97.55% 59.03% 100.00% 60.52%
554.roms_r 100.00% 73.79% 56.32% 100.00% 76.32%
Smaller binary sizes are always better.