SPEC CPU result comparison

15th October 2026 - Normalized report

SPEC2017 FPrate: Trunk LTO -O3 - generic vs x86-64-v3 vs native

-O3 -flto vs -O3 -flto and a given -march

AMD EPYC 8635P (zen5 based Sorano)

Run-times

Benchmark Generic x86-64-v3 Native x86-64-v3 Native
503.bwaves_r 100.00% 98.27% 98.27% 100.00% 100.00%
507.cactuBSSN_r 100.00% 80.41% 77.19% 100.00% 96.00%
508.namd_r 100.00% 89.60% 83.58% 100.00% 93.27%
510.parest_r 100.00% 98.12% 105.33% 100.00% 107.35%
511.povray_r 100.00% 98.75% 100.00% 100.00% 101.27%
519.lbm_r 100.00% 84.70% 79.00% 100.00% 93.28%
521.wrf_r 100.00% 89.20% 86.42% 100.00% 96.89%
526.blender_r 100.00% 99.30% 100.70% 100.00% 101.42%
527.cam4_r 100.00% 89.69% 90.41% 100.00% 100.80%
538.imagick_r 100.00% 76.88% 74.93% 100.00% 97.46%
544.nab_r 100.00% 93.56% 92.87% 100.00% 99.26%
549.fotonik3d_r 100.00% 103.42% 97.22% 100.00% 94.01%
554.roms_r 100.00% 76.97% 69.40% 100.00% 90.16%
The compiler used was GCC 17.0.0 20261015 (experimental) (revision 84348add236100e5f69a3e9ec4f7b1038bf2b83d).
Naturally, the smaller the run-times, the better.
See also continuous gcc tuning benchmarking at https://lnt.opensuse.org/db_default/v4/SPEC/spec_report/tuning.

Non-debug executable sizes

Benchmark Generic x86-64-v3 Native x86-64-v3 Native
503.bwaves_r 100.00% 95.61% 97.30% 100.00% 101.76%
507.cactuBSSN_r 100.00% 94.00% 101.33% 100.00% 107.80%
508.namd_r 100.00% 103.32% 103.64% 100.00% 100.31%
510.parest_r 100.00% 104.42% 107.08% 100.00% 102.54%
511.povray_r 100.00% 97.46% 99.09% 100.00% 101.67%
519.lbm_r 100.00% 93.94% 92.63% 100.00% 98.61%
521.wrf_r 100.00% 99.96% 103.85% 100.00% 103.89%
526.blender_r 100.00% 102.01% 103.60% 100.00% 101.55%
527.cam4_r 100.00% 100.15% 100.51% 100.00% 100.35%
538.imagick_r 100.00% 102.17% 106.94% 100.00% 104.67%
544.nab_r 100.00% 100.13% 101.70% 100.00% 101.57%
549.fotonik3d_r 100.00% 116.43% 117.89% 100.00% 101.25%
554.roms_r 100.00% 106.30% 106.77% 100.00% 100.44%
Smaller binary sizes are always better.

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

Run-times

Benchmark Generic x86-64-v3 Native x86-64-v3 Native
503.bwaves_r 100.00% 82.72% 82.72% 100.00% 100.00%
507.cactuBSSN_r 100.00% 87.05% 87.05% 100.00% 100.00%
508.namd_r 100.00% 86.96% 88.20% 100.00% 101.43%
510.parest_r 100.00% 96.98% 96.98% 100.00% 100.00%
511.povray_r 100.00% 96.84% 100.00% 100.00% 103.26%
519.lbm_r 100.00% 94.44% 93.80% 100.00% 99.32%
521.wrf_r 100.00% 87.85% 88.27% 100.00% 100.49%
526.blender_r 100.00% 97.66% 100.00% 100.00% 102.40%
527.cam4_r 100.00% 89.65% 89.65% 100.00% 100.00%
538.imagick_r 100.00% 72.77% 71.78% 100.00% 98.64%
544.nab_r 100.00% 93.48% 94.57% 100.00% 101.16%
549.fotonik3d_r 100.00% 96.46% 96.02% 100.00% 99.54%
554.roms_r 100.00% 78.12% 78.12% 100.00% 100.00%
The compiler used was GCC 17.0.0 20261015 (experimental) (revision 84348add236100e5f69a3e9ec4f7b1038bf2b83d).
Naturally, the smaller the run-times, the better.
See also continuous gcc tuning benchmarking at https://lnt.opensuse.org/db_default/v4/SPEC/spec_report/tuning.

Non-debug executable sizes

Benchmark Generic x86-64-v3 Native x86-64-v3 Native
503.bwaves_r 100.00% 95.62% 95.20% 100.00% 99.56%
507.cactuBSSN_r 100.00% 94.00% 92.99% 100.00% 98.93%
508.namd_r 100.00% 103.32% 103.21% 100.00% 99.89%
510.parest_r 100.00% 104.42% 104.94% 100.00% 100.50%
511.povray_r 100.00% 97.46% 97.83% 100.00% 100.37%
519.lbm_r 100.00% 94.19% 94.23% 100.00% 100.04%
521.wrf_r 100.00% 99.96% 103.20% 100.00% 103.24%
526.blender_r 100.00% 102.02% 102.37% 100.00% 100.34%
527.cam4_r 100.00% 100.15% 100.21% 100.00% 100.06%
538.imagick_r 100.00% 102.17% 102.94% 100.00% 100.75%
544.nab_r 100.00% 100.13% 100.35% 100.00% 100.23%
549.fotonik3d_r 100.00% 116.43% 116.14% 100.00% 99.75%
554.roms_r 100.00% 106.30% 105.73% 100.00% 99.46%
Smaller binary sizes are always better.