SPEC CPU result comparison

15th October 2026 - Normalized report

SPEC2017 INTrate: Trunk generic -O2 - standard vs LTO vs LTO+PGO

-O2

AMD EPYC 8635P (zen5 based Sorano)

Run-times

Benchmark Normal LTO LTO+PGO LTO LTO+PGO
500.perlbench_r 100.00% 99.34% 89.62% 100.00% 90.22%
502.gcc_r 100.00% 96.94% 90.41% 100.00% 93.26%
505.mcf_r 100.00% 98.59% 89.46% 100.00% 90.74%
520.omnetpp_r 100.00% 94.51% 78.46% 100.00% 83.01%
523.xalancbmk_r 100.00% 95.35% 91.54% 100.00% 96.01%
525.x264_r 100.00% 100.00% 77.60% 100.00% 77.60%
531.deepsjeng_r 100.00% 83.86% 91.70% 100.00% 109.36%
541.leela_r 100.00% 89.94% 85.71% 100.00% 95.30%
548.exchange2_r 100.00% 99.34% 65.71% 100.00% 66.15%
557.xz_r 100.00% 101.49% 98.08% 100.00% 96.64%
The compiler used was GCC 17.0.0 20261015 (experimental) (revision 84348add236100e5f69a3e9ec4f7b1038bf2b83d).
Naturally, the smaller the run-times, the better.
See also continuous gcc options benchmarking at https://lnt.opensuse.org/db_default/v4/SPEC/spec_report/options.

Non-debug executable sizes

Benchmark Normal LTO LTO+PGO LTO LTO+PGO
500.perlbench_r 100.00% 101.87% 86.26% 100.00% 84.68%
502.gcc_r 100.00% 100.94% 89.95% 100.00% 89.11%
505.mcf_r 100.00% 78.67% 74.97% 100.00% 95.30%
520.omnetpp_r 100.00% 85.31% 63.13% 100.00% 74.00%
523.xalancbmk_r 100.00% 86.93% 59.54% 100.00% 68.49%
525.x264_r 100.00% 99.68% 108.27% 100.00% 108.62%
531.deepsjeng_r 100.00% 99.08% 99.36% 100.00% 100.28%
541.leela_r 100.00% 60.53% 50.30% 100.00% 83.10%
548.exchange2_r 100.00% 87.23% 161.23% 100.00% 184.83%
557.xz_r 100.00% 68.25% 57.70% 100.00% 84.54%
Smaller binary sizes are always better.

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

Run-times

Benchmark LTO LTO+PGO
500.perlbench_r 100.00% 89.62%
502.gcc_r 100.00% 90.64%
505.mcf_r 100.00% 87.86%
520.omnetpp_r 100.00% 87.82%
523.xalancbmk_r 100.00% 96.21%
525.x264_r 100.00% 74.90%
531.deepsjeng_r 100.00% 105.34%
541.leela_r 100.00% 92.11%
548.exchange2_r 100.00% 56.39%
557.xz_r 100.00% 94.98%
The compiler used was GCC 17.0.0 20261015 (experimental) (revision 84348add236100e5f69a3e9ec4f7b1038bf2b83d).
Naturally, the smaller the run-times, the better.
See also continuous gcc options benchmarking at https://lnt.opensuse.org/db_default/v4/SPEC/spec_report/options.

Non-debug executable sizes

Benchmark LTO LTO+PGO
500.perlbench_r 100.00% 84.67%
502.gcc_r 100.00% 89.11%
505.mcf_r 100.00% 95.64%
520.omnetpp_r 100.00% 74.00%
523.xalancbmk_r 100.00% 68.50%
525.x264_r 100.00% 108.59%
531.deepsjeng_r 100.00% 100.28%
541.leela_r 100.00% 83.15%
548.exchange2_r 100.00% 184.80%
557.xz_r 100.00% 84.54%
Smaller binary sizes are always better.