We recently shared some benchmarks for Revolution R Open on the Windows platform, which showed significant improvements compared to R downloaded from CRAN. Those performance gains mainly come from multi-threading: Revolution R Open is linked to the Intel Math Kernel Library, which uses all available cores (rather than just one core) to compute matrix and vector operations in parallel.
The standard R build on Macs is already linked to a multithreaded math library, and if you build R yourself on Linux you can always link it to one. So it was interesting to see that Domino Data Labs benchmarked Revolution R Open against their stock build of R (already linked to a multithreaded library) and still found improved performance improvements: 30% faster on a single-core machine and 40% faster on a 4-core machine:
The Intel MKL is quite impressive in this regard: not only does it perform computations in parallel, but it also uses highly optimized algorithms and even optimizations at the machine-code level (pipelining operations). You can get those optimizations in Revolution R Open on all platforms, without changing a single line of R code.
Domino Data Lab blog: 40-percent faster R without any code changes
You should also benchmark against open-source R compiled against openblas.
Posted by: zach | November 11, 2014 at 11:16
@zach, here are some benchmarks comparing Revolution R Open with OpenBLAS.
Posted by: David Smith | November 11, 2014 at 14:20