SPEC CPU result comparison

15th October 2026 - Normalized report

SPEC2017 INTrate: 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
500.perlbench_r 100.00% 100.64% 95.97% 100.00% 95.37%
502.gcc_r 100.00% 101.71% 99.14% 100.00% 97.47%
505.mcf_r 100.00% 90.82% 90.32% 100.00% 99.45%
520.omnetpp_r 100.00% 97.76% 100.00% 100.00% 102.29%
523.xalancbmk_r 100.00% 85.71% 84.84% 100.00% 98.97%
525.x264_r 100.00% 90.59% 95.69% 100.00% 105.63%
531.deepsjeng_r 100.00% 93.22% 93.77% 100.00% 100.58%
541.leela_r 100.00% 102.20% 102.20% 100.00% 100.00%
548.exchange2_r 100.00% 92.02% 77.21% 100.00% 83.90%
557.xz_r 100.00% 99.57% 101.93% 100.00% 102.37%
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)
Naturally, the smaller the run-times, the better.

Non-debug executable sizes

Benchmark LLVM ICX GCC ICX GCC
500.perlbench_r 100.00% 115.39% 114.46% 100.00% 99.19%
502.gcc_r 100.00% 100.66% 101.04% 100.00% 100.37%
505.mcf_r 100.00% 95.08% 83.63% 100.00% 87.96%
520.omnetpp_r 100.00% 100.58% 99.69% 100.00% 99.11%
523.xalancbmk_r 100.00% 98.87% 103.63% 100.00% 104.81%
525.x264_r 100.00% 166.86% 111.82% 100.00% 67.02%
531.deepsjeng_r 100.00% 100.02% 102.34% 100.00% 102.31%
541.leela_r 100.00% 146.85% 106.91% 100.00% 72.80%
548.exchange2_r 100.00% 90.15% 18.62% 100.00% 20.66%
557.xz_r 100.00% 146.34% 112.86% 100.00% 77.12%
Smaller binary sizes are always better.

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

Run-times

Benchmark LLVM ICX GCC ICX GCC
500.perlbench_r 100.00% 95.16% 89.11% 100.00% 93.64%
502.gcc_r 100.00% 101.94% 97.63% 100.00% 95.77%
505.mcf_r 100.00% 92.15% 94.24% 100.00% 102.27%
520.omnetpp_r 100.00% 95.37% 93.83% 100.00% 98.38%
523.xalancbmk_r 100.00% 84.58% 89.80% 100.00% 106.17%
525.x264_r 100.00% 82.05% 87.82% 100.00% 107.03%
531.deepsjeng_r 100.00% 96.84% 95.63% 100.00% 98.75%
541.leela_r 100.00% 100.00% 102.13% 100.00% 102.13%
548.exchange2_r 100.00% 111.20% 92.95% 100.00% 83.58%
557.xz_r 100.00% 100.43% 103.43% 100.00% 102.99%
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)
Naturally, the smaller the run-times, the better.

Non-debug executable sizes

Benchmark LLVM ICX GCC ICX GCC
500.perlbench_r 100.00% 115.41% 115.33% 100.00% 99.93%
502.gcc_r 100.00% 101.88% 102.31% 100.00% 100.42%
505.mcf_r 100.00% 194.62% 80.77% 100.00% 41.50%
520.omnetpp_r 100.00% 101.77% 100.54% 100.00% 98.80%
523.xalancbmk_r 100.00% 100.49% 106.47% 100.00% 105.95%
525.x264_r 100.00% 168.26% 113.53% 100.00% 67.48%
531.deepsjeng_r 100.00% 100.19% 102.41% 100.00% 102.22%
541.leela_r 100.00% 164.00% 121.35% 100.00% 73.99%
548.exchange2_r 100.00% 91.06% 19.35% 100.00% 21.25%
557.xz_r 100.00% 143.55% 112.77% 100.00% 78.56%
Smaller binary sizes are always better.