2 ms·
I hate to post this multiple times, but the M1 has a dedicated matrix co-processor, it can do matrix multiplication at >1300GFLOP/s if you use the native Accele
by microtonal 5y ago
I hate to post this multiple times, but the M1 has a dedicated matrix co-processor, it can do matrix multiplication at >1300GFLOP/s if you use the native Accelerate framework (which uses the standard BLAS API). The M1 Pro/Max can even do double that (>2600 GFLOP/s) [1].
46.78 GFLOP/s is not even that great on non-specialized hardware. E.g., a Ryzen 5900X, can do ~150 GFLOP/s single-threaded with MKL.
[1] https://github.com/danieldk/gemm-benchmark#1-to-16-threads https://github.com/danieldk/gemm-benchmark#1-to-16-threads