SPEC CPU result comparison

15th October 2026 - Normalized report

SPEC2026 INTrate: Trunk generic -O2 - standard vs PGO vs LTO+PGO

-O2

AMD EPYC 8635P (zen5 based Sorano)

Run-times

Benchmark Normal PGO LTO+PGO PGO LTO+PGO
706.stockfish_r 100.00% 103.27% 101.31% 100.00% 98.10%
707.ntest_r 100.00% 100.63% 98.75% 100.00% 98.13%
708.sqlite_r 100.00% 94.12% 94.12% 100.00% 100.00%
710.omnetpp_r 100.00% 93.12% 75.00% 100.00% 80.54%
714.cpython_r 100.00% 81.96% 76.30% 100.00% 93.10%
721.gcc_r 100.00% 94.85% 90.31% 100.00% 95.22%
723.llvm_r 100.00% 94.51% 84.18% 100.00% 89.06%
727.cppcheck_r 100.00% 94.44% 88.03% 100.00% 93.21%
729.abc_r 100.00% 107.24% 109.05% 100.00% 101.69%
734.vpr_r 100.00% 98.92% 83.80% 100.00% 84.72%
735.gem5_r 100.00% 111.70% 109.40% 100.00% 97.95%
750.sealcrypto_r 100.00% 121.26% 123.19% 100.00% 101.59%
753.ns3_r 100.00% 100.00% 86.99% 100.00% 86.99%
777.zstd_r 100.00% 97.22% 97.22% 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
706.stockfish_r 100.00% 100.85% 86.90% 100.00% 86.17%
707.ntest_r 100.00% 95.05% 64.70% 100.00% 68.07%
708.sqlite_r 100.00% 89.75% 82.53% 100.00% 91.95%
710.omnetpp_r 100.00% 71.57% 59.66% 100.00% 83.37%
714.cpython_r 100.00% 88.48% 65.24% 100.00% 73.73%
721.gcc_r 100.00% 91.51% 84.30% 100.00% 92.12%
723.llvm_r 100.00% 65.48% 50.73% 100.00% 77.48%
727.cppcheck_r 100.00% 70.17% 59.84% 100.00% 85.28%
729.abc_r 100.00% 69.65% 49.28% 100.00% 70.75%
734.vpr_r 100.00% 74.25% 59.55% 100.00% 80.20%
735.gem5_r 100.00% 65.45% 50.85% 100.00% 77.68%
750.sealcrypto_r 100.00% 57.60% 33.18% 100.00% 57.61%
753.ns3_r 100.00% 67.46% 52.22% 100.00% 77.42%
777.zstd_r 100.00% 80.40% 67.78% 100.00% 84.31%
Smaller binary sizes are always better.

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

Run-times

Benchmark PGO LTO+PGO
706.stockfish_r 100.00% 97.10%
707.ntest_r 100.00% 96.50%
708.sqlite_r 100.00% 101.71%
710.omnetpp_r 100.00% 75.50%
714.cpython_r 100.00% 92.15%
721.gcc_r 100.00% 94.85%
723.llvm_r 100.00% 88.02%
727.cppcheck_r 100.00% 89.08%
729.abc_r 100.00% 100.63%
734.vpr_r 100.00% 84.41%
735.gem5_r 100.00% 96.77%
750.sealcrypto_r 100.00% 100.00%
753.ns3_r 100.00% 80.74%
777.zstd_r 100.00% 99.11%
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 PGO LTO+PGO
706.stockfish_r 100.00% 86.17%
707.ntest_r 100.00% 68.07%
708.sqlite_r 100.00% 91.96%
710.omnetpp_r 100.00% 83.37%
714.cpython_r 100.00% 73.73%
721.gcc_r 100.00% 92.12%
723.llvm_r 100.00% 77.49%
727.cppcheck_r 100.00% 85.29%
729.abc_r 100.00% 70.75%
734.vpr_r 100.00% 80.22%
735.gem5_r 100.00% 77.68%
750.sealcrypto_r 100.00% 57.60%
753.ns3_r 100.00% 77.42%
777.zstd_r 100.00% 84.31%
Smaller binary sizes are always better.