5 Responses

  1. Updated R & BLAS Timings – Mubashir Qasim

    […] article was first published on R – Strange Attractors, and kindly contributed to […]


  2. juan
    juan March 31, 2016 at 7:37 AM |

    Hello.
    What’s the difference between using your method (replacing openblas dll) vs using R Open from here https://mran.revolutionanalytics.com/rro/#intelmkl1 ?


    1. Avraham
      Avraham March 31, 2016 at 1:07 PM |

      Not much. I started before there was MS R, and the MKL was not available for use for commercial purposes, which is why I was originally restricted to ATLAS, Goto, and eventually OpenBLAS. Also, the OpenBLAS project has some studies showing OPB as fast if not faster than MKL, but I’ve never compared the two after building R. Since I maintain some packages for R on Windows, I need to be able to compile, so I’ve gotten used to building R from source. By all means, try them both and see which works better for you!


  3. Ricky
    Ricky May 19, 2016 at 7:17 PM |

    I compiled on i7 4790 (haswell) using -mtune=native , 64bit R.
    compared with original R, my Blas is about 10 times faster. NotBlas functions differs, A/B is like 3 times faster, simple calc is around 5% to 10% or no significant improvement. Overall, well done!

    have you compiled for 32bit as well? thanks.


    1. Avraham
      Avraham June 29, 2016 at 10:46 AM |

      Hi, @Ricky, thank you very much and sorry about the delay. Yes, I used to compile ATLAS for 32bit, as a matter of fact, the two latest entries at https://cran.r-project.org/bin/windows/contrib/ATLAS/ (C2D and C2i7) are my contributions. However, once I switched to a 64 bit operating system, I had severe issues compiling ATLAS, which is why I switched to OpenBLAS.


Comments are closed.