SPEC CPU result comparison

15th October 2026 - Normalized report

SPEC2017 INTrate: PGO+LTO -Ofast native - LLVM, ICX and GCC

-Ofast -native -flto -fprofile-instr-use vs -Ofast -march=native -flto -fprofile-use

AMD EPYC 8635P (zen5 based Sorano)

Run-times

Benchmark LLVM ICX GCC ICX GCC
500.perlbench_r 100.00% 97.51% 83.16% 100.00% 85.29%
502.gcc_r 100.00% 100.92% 97.47% 100.00% 96.58%
505.mcf_r 100.00% 89.78% 90.77% 100.00% 101.11%
520.omnetpp_r 100.00% 103.92% 88.48% 100.00% 85.14%
523.xalancbmk_r 100.00% 81.67% 88.19% 100.00% 107.98%
525.x264_r 100.00% 92.73% 95.16% 100.00% 102.61%
531.deepsjeng_r 100.00% 84.45% 90.43% 100.00% 107.08%
541.leela_r 100.00% 103.16% 105.58% 100.00% 102.35%
548.exchange2_r CE 100.00% 86.88%
557.xz_r 100.00% 101.89% 96.64% 100.00% 94.85%
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)
Naturally, the smaller the run-times, the better.

Non-debug executable sizes

Benchmark LLVM ICX GCC ICX GCC
500.perlbench_r 100.00% 112.70% 96.47% 100.00% 85.60%
502.gcc_r 100.00% 101.45% 91.52% 100.00% 90.21%
505.mcf_r 100.00% 223.15% 66.10% 100.00% 29.62%
520.omnetpp_r 100.00% 103.58% 82.51% 100.00% 79.66%
523.xalancbmk_r 100.00% 103.44% 82.70% 100.00% 79.95%
525.x264_r 100.00% 202.82% 91.58% 100.00% 45.16%
531.deepsjeng_r 100.00% 100.21% 102.09% 100.00% 101.88%
541.leela_r 100.00% 150.16% 78.33% 100.00% 52.17%
548.exchange2_r CE 100.00% 11.57%
557.xz_r 100.00% 164.34% 95.26% 100.00% 57.97%
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% 100.00% 81.94% 100.00% 81.94%
502.gcc_r 100.00% 100.71% 96.44% 100.00% 95.75%
505.mcf_r 100.00% 92.15% 95.29% 100.00% 103.41%
520.omnetpp_r 100.00% 100.91% 91.80% 100.00% 90.97%
523.xalancbmk_r 100.00% 86.70% 96.67% 100.00% 111.51%
525.x264_r 100.00% 85.22% 81.53% 100.00% 95.67%
531.deepsjeng_r 100.00% 101.04% 105.74% 100.00% 104.65%
541.leela_r 100.00% 103.40% 103.40% 100.00% 100.00%
548.exchange2_r CE 100.00% 84.33%
557.xz_r 100.00% 102.05% 93.85% 100.00% 91.97%
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% 112.93% 97.14% 100.00% 86.02%
502.gcc_r 100.00% 101.66% 92.11% 100.00% 90.61%
505.mcf_r 100.00% 207.35% 63.30% 100.00% 30.53%
520.omnetpp_r 100.00% 104.99% 83.74% 100.00% 79.76%
523.xalancbmk_r 100.00% 105.01% 84.32% 100.00% 80.29%
525.x264_r 100.00% 207.84% 94.14% 100.00% 45.30%
531.deepsjeng_r 100.00% 100.67% 102.31% 100.00% 101.63%
541.leela_r 100.00% 161.75% 86.26% 100.00% 53.33%
548.exchange2_r CE 100.00% 11.73%
557.xz_r 100.00% 161.18% 94.09% 100.00% 58.38%
Smaller binary sizes are always better.