Benchmarks

Performance of Eigen’s dense kernels measured against BLAS/LAPACK libraries, one machine and one dataset per page, each with the conditions it was taken under: CPU, operating system, compiler and flags, Eigen commit, library versions, thread count and method. Ratios between Eigen and a library are formed on one machine with one build configuration; absolute rates are not comparable across pages.

Current results

Intel Xeon Gold 6444Y, September 2026: twelve BLAS/LAPACK operations in float and double against OpenBLAS 0.3.26 and native Intel oneMKL 2025.3, with upstream BLIS 2.1 for the six BLAS operations; single thread, AVX-512, Eigen master ab4f7316d.
AMD Zen 5 (Ryzen AI MAX+ 395), September 2026: AXPY, DOT, GEMV, GEMM, TRSM, SYRK, POTRF, GETRF, GEQRF, SYEV, GESDD and GEEV in float and double against OpenBLAS 0.3.26, Intel oneMKL 2025.3 and AMD AOCL 5.2, single thread, AVX-512, Eigen master 15f227178.
AMD Zen 5 (Ryzen AI MAX+ 395) limited to AVX2, September 2026: the same operations, shapes and libraries with Eigen built for AVX2 and FMA and every library held to its AVX2 kernels, Eigen master 15f227178.
Apple M4 (Mac mini) with SME, September 2026: the same operations and shapes against Apple Accelerate (macOS 27.0), Arm Performance Libraries 26.07 and OpenBLAS 0.3.34, single thread, Eigen master 46a2d0c56 built with -mcpu=apple-m4, which selects its SME matrix-product kernel.
Apple M4 (Mac mini) with NEON, September 2026: the same cells with Eigen built for NEON alone, against OpenBLAS, Eigen master 46a2d0c56. Accelerate and Arm Performance Libraries cannot be held to NEON through documented means; the SME page has those comparisons.

Every page is produced by the comparison harness under benchmarks/comparison/ in the Eigen source tree (in review as merge request !2903), which validates each reference library’s result against Eigen before timing it and records the full provenance of a run. The charts are drawn from the raw result files by scripts/benchmarks/plot_comparison.py in the website repository, where the result files behind each page are kept beside its images. The plan and progress are tracked in Eigen issue #3112.

Historical results (2009–2011)

These pages measured Eigen 2 and an early Eigen 3 with the BTL harness, which has since been removed from the tree, against libraries that have all been superseded. They are kept as a record; the numbers say nothing about current hardware or current Eigen.

Benchmark — March 2011, Core 2 Quad Q9400, Eigen 3 and 2 vs MKL 11.0, GotoBLAS 2, ATLAS 3.8.3, ACML 4.2.0
Benchmark Intel — February 2009, Pentium D, Eigen 2
Benchmark AMD — February 2009, Athlon 64 X2, Eigen 2
Benchmark AMD Intel compare — February 2009

Eigen vs. BLAS/LAPACK libraries, AMD Zen 5 (September 2026)

Single-threaded BLAS (AXPY, DOT, GEMV, GEMM, TRSM, SYRK) and LAPACK (POTRF, GETRF, GEQRF, SYEV, GESDD, GEEV): Eigen master against OpenBLAS, Intel oneMKL and AMD AOCL on an AMD Ryzen AI MAX+ 395.

Eigen vs. BLAS/LAPACK libraries, AMD Zen 5 limited to AVX2 (September 2026)

Single-threaded BLAS (AXPY, DOT, GEMV, GEMM, TRSM, SYRK) and LAPACK (POTRF, GETRF, GEQRF, SYEV, GESDD, GEEV): Eigen master against OpenBLAS, Intel oneMKL and AMD AOCL on an AMD Ryzen AI MAX+ 395, every library limited to AVX2 and FMA.

Eigen vs. BLAS/LAPACK libraries, Intel Xeon Gold 6444Y (September 2026)

Single-threaded float and double BLAS/LAPACK comparisons on an Intel Xeon Gold 6444Y: Eigen against OpenBLAS and oneMKL, with upstream BLIS for BLAS.

Eigen vs. BLAS/LAPACK libraries, Apple M4 with SME (September 2026)

Single-threaded BLAS (AXPY, DOT, GEMV, GEMM, TRSM, SYRK) and LAPACK (POTRF, GETRF, GEQRF, SYEV, GESDD, GEEV): Eigen master, built for the Apple M4 with its SME matrix-product kernel, against Apple Accelerate, Arm Performance Libraries and OpenBLAS on a Mac mini.