Parallel Computing for Data Science

12.5. EXAMPLE: PERFORMANCE OF OPENBLAS 261

Thus even in ordinary serial computation, e.g. R’s %*% operator, the speed

of matrix operations may vary according to the version of BLAS that your

implementation of R was built with.

In our context here of parallel computation, one version of BLAS of spe-

cial interest is OpenBLAS, a multithreaded version that can bring perfor-

mance gains on multicore machines, and also includes various efficiencies

that greatly improve performance even in the serial case. We’ll take a closer

look at it in Section 12.5.

As noted, there is also CUBLAS, a version of BLAS for NVIDIA GPUs,

highly tailored to that platform. A num ...