SPEC CPU result comparison

15th October 2026 - Normalized report

SPEC2026 INTrate: Trunk native -Ofast - standard vs PGO vs LTO+PGO

-Ofast -march=native -mtune=native

Please note that even though the header of this comparison says -Ofast, benchmarks 710.omnetpp_r, 734.vpr_r, 735.gem5_r and 753.ns3_r were additionally compiled with -fno-finite-math-only.

AMD EPYC 8635P (zen5 based Sorano)

Run-times

Benchmark Normal PGO LTO+PGO PGO LTO+PGO
706.stockfish_r 100.00% 103.81% 99.05% 100.00% 95.41%
707.ntest_r 100.00% 100.00% 94.58% 100.00% 94.58%
708.sqlite_r 100.00% 93.68% 93.68% 100.00% 100.00%
710.omnetpp_r 100.00% 88.99% 68.12% 100.00% 76.55%
714.cpython_r 100.00% 80.87% 74.13% 100.00% 91.67%
721.gcc_r 100.00% 92.74% 86.93% 100.00% 93.74%
723.llvm_r 100.00% 90.45% 84.08% 100.00% 92.96%
727.cppcheck_r 100.00% 100.00% 84.11% 100.00% 84.11%
729.abc_r 100.00% 106.39% 105.48% 100.00% 99.14%
734.vpr_r 100.00% 97.05% 84.13% 100.00% 86.68%
735.gem5_r 100.00% 106.50% 103.25% 100.00% 96.95%
750.sealcrypto_r 100.00% 116.67% 116.67% 100.00% 100.00%
753.ns3_r 100.00% 96.86% 83.47% 100.00% 86.18%
777.zstd_r 100.00% 102.17% 101.45% 100.00% 99.29%
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% 97.01% 83.69% 100.00% 86.27%
707.ntest_r 100.00% 93.79% 63.91% 100.00% 68.14%
708.sqlite_r 100.00% 73.45% 65.67% 100.00% 89.42%
710.omnetpp_r 100.00% 69.51% 58.46% 100.00% 84.10%
714.cpython_r 100.00% 80.68% 59.79% 100.00% 74.11%
721.gcc_r 100.00% 85.86% 82.47% 100.00% 96.05%
723.llvm_r 100.00% 56.09% 44.13% 100.00% 78.69%
727.cppcheck_r 100.00% 73.90% 64.87% 100.00% 87.79%
729.abc_r 100.00% 50.45% 35.85% 100.00% 71.07%
734.vpr_r 100.00% 71.66% 58.25% 100.00% 81.29%
735.gem5_r 100.00% 60.87% 48.40% 100.00% 79.51%
750.sealcrypto_r 100.00% 47.37% 26.88% 100.00% 56.74%
753.ns3_r 100.00% 67.10% 54.12% 100.00% 80.66%
777.zstd_r 100.00% 49.42% 41.35% 100.00% 83.68%
Smaller binary sizes are always better.

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

Run-times

Benchmark Normal PGO LTO+PGO PGO LTO+PGO
706.stockfish_r 100.00% 102.01% 102.01% 100.00% 100.00%
707.ntest_r 100.00% 96.76% 91.20% 100.00% 94.26%
708.sqlite_r 100.00% 91.42% 90.26% 100.00% 98.73%
710.omnetpp_r 100.00% 88.84% 64.42% 100.00% 72.51%
714.cpython_r 100.00% 78.21% 71.68% 100.00% 91.64%
721.gcc_r 100.00% 91.61% 84.70% 100.00% 92.45%
723.llvm_r 100.00% 86.37% 76.10% 100.00% 88.11%
727.cppcheck_r 100.00% 95.10% 88.70% 100.00% 93.27%
729.abc_r 100.00% 105.73% 105.05% 100.00% 99.35%
734.vpr_r 100.00% 97.84% 83.84% 100.00% 85.68%
735.gem5_r 100.00% 108.77% 104.50% 100.00% 96.08%
750.sealcrypto_r 100.00% 105.23% 103.49% 100.00% 98.34%
753.ns3_r 100.00% 88.59% 69.50% 100.00% 78.45%
777.zstd_r 100.00% 96.40% 95.06% 100.00% 98.60%
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% 97.05% 83.68% 100.00% 86.22%
707.ntest_r 100.00% 94.52% 64.48% 100.00% 68.22%
708.sqlite_r 100.00% 71.04% 63.68% 100.00% 89.65%
710.omnetpp_r 100.00% 69.55% 58.63% 100.00% 84.30%
714.cpython_r 100.00% 80.54% 59.67% 100.00% 74.09%
721.gcc_r 100.00% 86.17% 82.75% 100.00% 96.03%
723.llvm_r 100.00% 57.29% 45.16% 100.00% 78.82%
727.cppcheck_r 100.00% 74.24% 65.15% 100.00% 87.76%
729.abc_r 100.00% 50.75% 36.07% 100.00% 71.07%
734.vpr_r 100.00% 71.89% 58.46% 100.00% 81.32%
735.gem5_r 100.00% 61.40% 48.81% 100.00% 79.49%
750.sealcrypto_r 100.00% 47.68% 27.25% 100.00% 57.15%
753.ns3_r 100.00% 67.54% 54.49% 100.00% 80.67%
777.zstd_r 100.00% 51.07% 42.79% 100.00% 83.79%
Smaller binary sizes are always better.