OpenBLAS

A developer has benchmarked matrix multiplication in Intel oneAPI 2026.1 against NumPy with OpenBLAS, running on a Core i7-1195G7 with integrated Iris Xe graphics. Plain DPC++ code ran about twice as fast as NumPy/OpenBLAS at 0.062 seconds versus 0.124, and adding cache optimizations brought the time down to 0.048 seconds, roughly 2.58 times faster. The results suggest Intel's SYCL toolchain can meaningfully outperform the standard CPU BLAS path on supported Intel hardware, even on integrated GPUs.