SPEC CPU result comparison

15th October 2026 - Normalized report

SPEC2017 INTrate: Trunk native -Ofast - standard vs PGO vs LTO+PGO

-Ofast -march=native -mtune=native

AMD EPYC 8635P (zen5 based Sorano)

Run-times

Benchmark Normal PGO LTO+PGO PGO LTO+PGO
500.perlbench_r 100.00% 88.63% 85.84% 100.00% 96.85%
502.gcc_r 100.00% 94.95% 89.26% 100.00% 94.01%
505.mcf_r 100.00% 99.06% 85.25% 100.00% 86.05%
520.omnetpp_r 100.00% 94.23% 83.37% 100.00% 88.48%
523.xalancbmk_r 100.00% 118.07% 107.18% 100.00% 90.78%
525.x264_r 100.00% 101.15% 104.96% 100.00% 103.77%
531.deepsjeng_r 100.00% 111.19% 90.00% 100.00% 80.94%
541.leela_r 100.00% 103.76% 96.24% 100.00% 92.75%
548.exchange2_r 100.00% 103.91% 98.93% 100.00% 95.21%
557.xz_r 100.00% 95.63% 95.63% 100.00% 100.00%
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 PGO LTO+PGO PGO LTO+PGO
500.perlbench_r 100.00% 78.46% 75.93% 100.00% 96.78%
502.gcc_r 100.00% 87.99% 90.39% 100.00% 102.72%
505.mcf_r 100.00% 71.19% 57.77% 100.00% 81.15%
520.omnetpp_r 100.00% 70.85% 59.15% 100.00% 83.49%
523.xalancbmk_r 100.00% 58.44% 46.53% 100.00% 79.62%
525.x264_r 100.00% 86.61% 88.27% 100.00% 101.92%
531.deepsjeng_r 100.00% 100.30% 99.17% 100.00% 98.88%
541.leela_r 100.00% 67.98% 40.45% 100.00% 59.51%
548.exchange2_r 100.00% 83.88% 49.21% 100.00% 58.67%
557.xz_r 100.00% 82.16% 53.91% 100.00% 65.62%
Smaller binary sizes are always better.

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

Run-times

Benchmark Normal PGO LTO+PGO PGO LTO+PGO
500.perlbench_r 100.00% 87.89% 83.41% 100.00% 94.90%
502.gcc_r 100.00% 93.15% 86.94% 100.00% 93.33%
505.mcf_r 100.00% 96.98% 84.45% 100.00% 87.08%
520.omnetpp_r 100.00% 98.06% 86.67% 100.00% 88.38%
523.xalancbmk_r 100.00% 115.19% 101.87% 100.00% 88.44%
525.x264_r 100.00% 103.06% 105.10% 100.00% 101.98%
531.deepsjeng_r 100.00% 106.95% 90.81% 100.00% 84.91%
541.leela_r 100.00% 99.59% 87.12% 100.00% 87.47%
548.exchange2_r 100.00% 102.14% 96.58% 100.00% 94.56%
557.xz_r 100.00% 92.59% 94.24% 100.00% 101.78%
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 PGO LTO+PGO PGO LTO+PGO
500.perlbench_r 100.00% 78.66% 76.24% 100.00% 96.91%
502.gcc_r 100.00% 88.30% 90.75% 100.00% 102.77%
505.mcf_r 100.00% 75.99% 61.21% 100.00% 80.54%
520.omnetpp_r 100.00% 71.07% 59.35% 100.00% 83.52%
523.xalancbmk_r 100.00% 57.20% 45.49% 100.00% 79.53%
525.x264_r 100.00% 86.04% 87.79% 100.00% 102.03%
531.deepsjeng_r 100.00% 100.37% 99.24% 100.00% 98.87%
541.leela_r 100.00% 67.43% 40.42% 100.00% 59.94%
548.exchange2_r 100.00% 85.70% 48.18% 100.00% 56.22%
557.xz_r 100.00% 81.40% 53.46% 100.00% 65.68%
Smaller binary sizes are always better.