SPEC CPU result comparison

15th October 2026 - Normalized report

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

-Ofast -march=native -flto

AMD EPYC 8635P (zen5 based Sorano)

Run-times

Benchmark LLVM ICX GCC ICX GCC
503.bwaves_r 100.00% 87.33% 81.00% 100.00% 92.75%
507.cactuBSSN_r 100.00% 87.97% 96.99% 100.00% 110.26%
508.namd_r 100.00% 97.57% 107.57% 100.00% 110.25%
510.parest_r 100.00% 86.01% 80.77% 100.00% 93.90%
511.povray_r 100.00% 92.03% 96.62% 100.00% 104.99%
519.lbm_r 100.00% 42.92% 49.65% 100.00% 115.68%
521.wrf_r 100.00% 88.24% 97.06% 100.00% 110.00%
526.blender_r 100.00% 94.00% 100.72% 100.00% 107.14%
527.cam4_r 100.00% 71.09% 78.04% 100.00% 109.79%
538.imagick_r 100.00% 129.13% 52.75% 100.00% 40.85%
544.nab_r 100.00% 91.01% 104.35% 100.00% 114.65%
549.fotonik3d_r 100.00% 109.78% 100.67% 100.00% 91.70%
554.roms_r 100.00% 233.33% 93.43% 100.00% 40.04%
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 LLVM ICX GCC ICX GCC
503.bwaves_r 100.00% 82.08% 4.44% 100.00% 5.41%
507.cactuBSSN_r 100.00% 121.95% 97.59% 100.00% 80.02%
508.namd_r 100.00% 108.83% 122.83% 100.00% 112.86%
510.parest_r 100.00% 109.08% 90.52% 100.00% 82.98%
511.povray_r 100.00% 134.00% 90.40% 100.00% 67.46%
519.lbm_r 100.00% 125.70% 52.10% 100.00% 41.45%
521.wrf_r 100.00% 92.78% 96.19% 100.00% 103.68%
526.blender_r 100.00% 84.11% 79.61% 100.00% 94.65%
527.cam4_r 100.00% 112.71% 100.81% 100.00% 89.44%
538.imagick_r 100.00% 178.78% 121.31% 100.00% 67.86%
544.nab_r 100.00% 162.45% 102.04% 100.00% 62.81%
549.fotonik3d_r 100.00% 90.73% 34.80% 100.00% 38.35%
554.roms_r 100.00% 85.28% 58.44% 100.00% 68.53%
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% 84.13% 84.13% 100.00% 100.00%
507.cactuBSSN_r 100.00% 86.47% 91.98% 100.00% 106.38%
508.namd_r 100.00% 104.93% 105.67% 100.00% 100.70%
510.parest_r 100.00% 97.70% 95.79% 100.00% 98.04%
511.povray_r 100.00% 93.36% 97.79% 100.00% 104.74%
519.lbm_r 100.00% 130.17% 160.34% 100.00% 123.18%
521.wrf_r 100.00% 84.31% 95.24% 100.00% 112.96%
526.blender_r 100.00% 90.99% 111.49% 100.00% 122.52%
527.cam4_r 100.00% 76.03% 82.57% 100.00% 108.60%
538.imagick_r 100.00% 108.63% 69.01% 100.00% 63.53%
544.nab_r 100.00% 103.99% 106.55% 100.00% 102.47%
549.fotonik3d_r 100.00% 101.87% 100.47% 100.00% 98.62%
554.roms_r 100.00% 133.67% 92.67% 100.00% 69.33%
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% 82.69% 4.64% 100.00% 5.61%
507.cactuBSSN_r 100.00% 145.29% 95.31% 100.00% 65.60%
508.namd_r 100.00% 117.27% 135.48% 100.00% 115.53%
510.parest_r 100.00% 112.78% 93.41% 100.00% 82.82%
511.povray_r 100.00% 141.85% 96.09% 100.00% 67.74%
519.lbm_r 100.00% 155.13% 71.09% 100.00% 45.82%
521.wrf_r 100.00% 102.14% 101.01% 100.00% 98.89%
526.blender_r 100.00% 85.55% 81.83% 100.00% 95.65%
527.cam4_r 100.00% 114.71% 101.49% 100.00% 88.47%
538.imagick_r 100.00% 192.74% 115.32% 100.00% 59.83%
544.nab_r 100.00% 162.71% 101.50% 100.00% 62.38%
549.fotonik3d_r 100.00% 94.70% 34.94% 100.00% 36.90%
554.roms_r 100.00% 92.02% 59.13% 100.00% 64.26%
Smaller binary sizes are always better.