SPEC CPU result comparison

15th October 2026 - Normalized report

SPEC2017 FPrate: Trunk LTO -Ofast - previous znver's vs native

-Ofast -flto -march=znverX vs -Ofast -flto -march=native

AMD EPYC 8635P (zen5 based Sorano)

Run-times

Benchmark Znver3 Znver4 Native Znver4 Native
503.bwaves_r 100.00% 100.00% 100.00% 100.00% 100.00%
507.cactuBSSN_r 100.00% 94.91% 93.82% 100.00% 98.85%
508.namd_r 100.00% 93.21% 93.21% 100.00% 100.00%
510.parest_r 100.00% 97.47% 97.47% 100.00% 100.00%
511.povray_r 100.00% 97.09% 97.09% 100.00% 100.00%
519.lbm_r 100.00% 88.66% 89.92% 100.00% 101.42%
521.wrf_r 100.00% 97.88% 97.88% 100.00% 100.00%
526.blender_r 100.00% 93.96% 93.96% 100.00% 100.00%
527.cam4_r 100.00% 104.06% 104.06% 100.00% 100.00%
538.imagick_r 100.00% 87.17% 87.17% 100.00% 100.00%
544.nab_r 100.00% 100.00% 100.00% 100.00% 100.00%
549.fotonik3d_r 100.00% 97.30% 94.18% 100.00% 96.79%
554.roms_r 100.00% 90.45% 90.45% 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 Znver3 Znver4 Native Znver4 Native
503.bwaves_r 100.00% 97.60% 96.51% 100.00% 98.88%
507.cactuBSSN_r 100.00% 109.25% 108.76% 100.00% 99.55%
508.namd_r 100.00% 97.96% 103.46% 100.00% 105.62%
510.parest_r 100.00% 102.28% 102.40% 100.00% 100.13%
511.povray_r 100.00% 101.21% 101.33% 100.00% 100.11%
519.lbm_r 100.00% 94.85% 95.20% 100.00% 100.37%
521.wrf_r 100.00% 99.30% 99.40% 100.00% 100.10%
526.blender_r 100.00% 101.09% 101.37% 100.00% 100.28%
527.cam4_r 100.00% 100.14% 100.16% 100.00% 100.02%
538.imagick_r 100.00% 107.32% 107.60% 100.00% 100.26%
544.nab_r 100.00% 100.89% 101.25% 100.00% 100.35%
549.fotonik3d_r 100.00% 103.82% 102.44% 100.00% 98.67%
554.roms_r 100.00% 100.52% 100.74% 100.00% 100.22%
Smaller binary sizes are always better.