SPEC CPU result comparison

15th October 2026 - Normalized report

SPEC2017 FPrate: ThinLTO and LTO native -Ofast - LLVM and GCC

-Ofast -march=native -flto=thin vs. -Ofast -march=native -flto

AMD EPYC 8635P (zen5 based Sorano)

Run-times

Benchmark LLVM ThinLTO LLVM FullLTO GCC LLVM FullLTO GCC
503.bwaves_r 100.00% 89.11% 72.18% 100.00% 81.00%
507.cactuBSSN_r 100.00% 100.00% 96.99% 100.00% 96.99%
508.namd_r 100.00% 98.93% 106.42% 100.00% 107.57%
510.parest_r 100.00% 100.00% 80.77% 100.00% 80.77%
511.povray_r 100.00% 102.22% 98.77% 100.00% 96.62%
519.lbm_r 100.00% 99.31% 49.31% 100.00% 49.65%
521.wrf_r 100.00% 95.20% 92.40% 100.00% 97.06%
526.blender_r 100.00% 100.97% 101.69% 100.00% 100.72%
527.cam4_r 100.00% 99.14% 77.37% 100.00% 78.04%
538.imagick_r 100.00% 99.04% 52.24% 100.00% 52.75%
544.nab_r 100.00% 97.46% 101.69% 100.00% 104.35%
549.fotonik3d_r 100.00% 98.90% 99.56% 100.00% 100.67%
554.roms_r 100.00% 99.07% 92.56% 100.00% 93.43%
LLVM clang version 23.1.3 (/labs/mjambor/git/llvm 0d261d1ca552c95a8f007e061c787ac7132fbcbc), flang version 23.1.3 (/labs/mjambor/git/llvm 0d261d1ca552c95a8f007e061c787ac7132fbcbc)
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 ThinLTO LLVM FullLTO GCC LLVM FullLTO GCC
503.bwaves_r 100.00% 100.78% 4.48% 100.00% 4.44%
507.cactuBSSN_r 100.00% 97.92% 95.55% 100.00% 97.59%
508.namd_r 100.00% 86.87% 106.70% 100.00% 122.83%
510.parest_r 100.00% 92.07% 83.34% 100.00% 90.52%
511.povray_r 100.00% 106.77% 96.51% 100.00% 90.40%
519.lbm_r 100.00% 85.93% 44.77% 100.00% 52.10%
521.wrf_r 100.00% 99.59% 95.80% 100.00% 96.19%
526.blender_r 100.00% 95.75% 76.22% 100.00% 79.61%
527.cam4_r 100.00% 98.19% 98.99% 100.00% 100.81%
538.imagick_r 100.00% 94.73% 114.92% 100.00% 121.31%
544.nab_r 100.00% 97.21% 99.19% 100.00% 102.04%
549.fotonik3d_r 100.00% 106.38% 37.02% 100.00% 34.80%
554.roms_r 100.00% 91.88% 53.70% 100.00% 58.44%
Smaller binary sizes are always better.

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

Run-times

Benchmark LLVM ThinLTO LLVM FullLTO GCC LLVM FullLTO GCC
503.bwaves_r 100.00% 79.35% 66.75% 100.00% 84.13%
507.cactuBSSN_r 100.00% 99.50% 91.52% 100.00% 91.98%
508.namd_r 100.00% 95.31% 100.70% 100.00% 105.67%
510.parest_r 100.00% 98.86% 94.70% 100.00% 95.79%
511.povray_r 100.00% 102.26% 100.00% 100.00% 97.79%
519.lbm_r 100.00% 100.00% 160.34% 100.00% 160.34%
521.wrf_r 100.00% 100.00% 95.24% 100.00% 95.24%
526.blender_r 100.00% 100.45% 111.99% 100.00% 111.49%
527.cam4_r 100.00% 98.29% 81.16% 100.00% 82.57%
538.imagick_r 100.00% 99.37% 68.57% 100.00% 69.01%
544.nab_r 100.00% 95.12% 101.36% 100.00% 106.55%
549.fotonik3d_r 100.00% 100.00% 100.47% 100.00% 100.47%
554.roms_r 100.00% 98.04% 90.85% 100.00% 92.67%
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)
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 ThinLTO LLVM FullLTO GCC LLVM FullLTO GCC
503.bwaves_r 100.00% 100.03% 4.64% 100.00% 4.64%
507.cactuBSSN_r 100.00% 97.24% 92.68% 100.00% 95.31%
508.namd_r 100.00% 83.87% 113.62% 100.00% 135.48%
510.parest_r 100.00% 91.93% 85.86% 100.00% 93.41%
511.povray_r 100.00% 102.99% 98.96% 100.00% 96.09%
519.lbm_r 100.00% 83.03% 59.02% 100.00% 71.09%
521.wrf_r 100.00% 101.30% 102.33% 100.00% 101.01%
526.blender_r 100.00% 95.01% 77.75% 100.00% 81.83%
527.cam4_r 100.00% 98.69% 100.16% 100.00% 101.49%
538.imagick_r 100.00% 96.50% 111.28% 100.00% 115.32%
544.nab_r 100.00% 98.01% 99.48% 100.00% 101.50%
549.fotonik3d_r 100.00% 104.92% 36.66% 100.00% 34.94%
554.roms_r 100.00% 92.67% 54.80% 100.00% 59.13%
Smaller binary sizes are always better.