SPEC CPU result comparison

15th October 2026 - Normalized report

SPEC2017 FPrate: Standard -Ofast native - LLVM, ICX and GCC

-Ofast -march=native

AMD EPYC 8635P (zen5 based Sorano)

Run-times

Benchmark ICX GCC
503.bwaves_r 100.00% 90.59%
507.cactuBSSN_r 100.00% 101.98%
508.namd_r 100.00% 110.25%
510.parest_r 100.00% 96.00%
511.povray_r 100.00% 96.22%
519.lbm_r 100.00% 93.78%
521.wrf_r 100.00% 106.45%
526.blender_r 100.00% 107.09%
527.cam4_r 100.00% 108.28%
538.imagick_r 100.00% 42.07%
544.nab_r 100.00% 105.26%
549.fotonik3d_r 100.00% 95.17%
554.roms_r 100.00% 40.85%
LLVM clang version 23.1.3 (/labs/mjambor/git/llvm 0d261d1ca552c95a8f007e061c787ac7132fbcbc), flang version 23.1.3 (/labs/mjambor/git/llvm 0d261d1ca552c95a8f007e061c787ac7132fbcbc)
ICX Intel(R) oneAPI DPC++/C++ Compiler 2026.1.1 (2026.1.1.20260724), ifx (IFX) 2026.1.1 20260724
GCC GCC 17.0.0 20261015 (experimental) (revision 84348add236100e5f69a3e9ec4f7b1038bf2b83d)
In the case of LLVM -fno-finite-math-only was used to compile 511.povray_r (that benchmark only) to avoid a known issue.
Naturally, the smaller the run-times, the better.

Non-debug executable sizes

Benchmark ICX GCC
503.bwaves_r 100.00% 5.35%
507.cactuBSSN_r 100.00% 89.32%
508.namd_r 100.00% 89.44%
510.parest_r 100.00% 101.52%
511.povray_r 100.00% 69.46%
519.lbm_r 100.00% 91.43%
521.wrf_r 100.00% 96.33%
526.blender_r 100.00% 93.87%
527.cam4_r 100.00% 156.99%
538.imagick_r 100.00% 84.59%
544.nab_r 100.00% 77.55%
549.fotonik3d_r 100.00% 38.50%
554.roms_r 100.00% 72.64%
Smaller binary sizes are always better.

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

Run-times

Benchmark LLVM ICX GCC ICX GCC
503.bwaves_r 100.00% 81.09% 56.54% 100.00% 69.73%
507.cactuBSSN_r 100.00% 90.48% 90.98% 100.00% 100.55%
508.namd_r 100.00% 103.42% 104.89% 100.00% 101.42%
510.parest_r 100.00% 87.83% 84.87% 100.00% 96.63%
511.povray_r 100.00% 96.36% 97.17% 100.00% 100.84%
519.lbm_r 100.00% 173.19% 150.21% 100.00% 86.73%
521.wrf_r 100.00% 87.19% 95.26% 100.00% 109.27%
526.blender_r 100.00% 91.54% 107.35% 100.00% 117.27%
527.cam4_r 100.00% 77.13% 82.33% 100.00% 106.74%
538.imagick_r 100.00% 98.30% 63.35% 100.00% 64.45%
544.nab_r 100.00% 104.36% 103.27% 100.00% 98.96%
549.fotonik3d_r 100.00% 103.32% 101.42% 100.00% 98.17%
554.roms_r 100.00% 133.01% 91.50% 100.00% 68.80%
LLVM clang version 23.1.3 (https://github.com/llvm/llvm-project.git 0d261d1ca552c95a8f007e061c787ac7132fbcbc), flang version 23.1.3 (https://github.com/llvm/llvm-project.git 0d261d1ca552c95a8f007e061c787ac7132fbcbc)
ICX Intel(R) oneAPI DPC++/C++ Compiler 2026.1.1 (2026.1.1.20260724), ifx (IFX) 2026.1.1 20260724
GCC GCC 17.0.0 20261015 (experimental) (revision 84348add236100e5f69a3e9ec4f7b1038bf2b83d)
In the case of LLVM -fno-finite-math-only was used to compile 511.povray_r (that benchmark only) to avoid a known issue.
Naturally, the smaller the run-times, the better.

Non-debug executable sizes

Benchmark LLVM ICX GCC ICX GCC
503.bwaves_r 100.00% 82.87% 4.55% 100.00% 5.49%
507.cactuBSSN_r 100.00% 138.66% 102.08% 100.00% 73.62%
508.namd_r 100.00% 121.92% 116.66% 100.00% 95.68%
510.parest_r 100.00% 102.03% 99.83% 100.00% 97.84%
511.povray_r 100.00% 157.08% 109.25% 100.00% 69.55%
519.lbm_r 100.00% 64.23% 68.82% 100.00% 107.16%
521.wrf_r 100.00% 106.46% 100.32% 100.00% 94.22%
526.blender_r 100.00% 91.27% 86.56% 100.00% 94.85%
527.cam4_r 100.00% 154.75% 239.24% 100.00% 154.60%
538.imagick_r 100.00% 188.69% 142.37% 100.00% 75.45%
544.nab_r 100.00% 128.19% 99.33% 100.00% 77.49%
549.fotonik3d_r 100.00% 99.94% 37.22% 100.00% 37.24%
554.roms_r 100.00% 90.65% 63.69% 100.00% 70.26%
Smaller binary sizes are always better.