Hi, I want to create a custom build of R, for my particular machine that will run as fast as possible. CPU: Dual cora AMD 64 bit OS: Fedora 15 I know there are a lot of options for compile time flags, lapack, blas, atlas, etc. Not sure what the best combination is. Has someone already worked this out? Thanks! -- Noah Silverman UCLA Department of Statistics 8117 Math Sciences Building Los Angeles, CA 90095
On Sun, Aug 21, 2011 at 10:18 AM, Noah Silverman <noahsilverman at ucla.edu> wrote:> Hi, > > I want to create a custom build of R, for my particular machine that will run as fast as possible. > > CPU: Dual cora AMD 64 bit > > OS: Fedora 15 > > I know there are a lot of options for compile time flags, lapack, blas, atlas, etc. ?Not sure what the best combination is.If you need any kind of matrix manipulations with larger matrices (say more than a few hundred rows or columns), I would definitely suggest a custom BLAS, for example Goto BLAS or ATLAS. When I last compared them a couple of years ago, Goto BLAS was somewhat (on the order of 20%) faster than ATLAS. This may have changed because of better compilers, plus I only used Intel processors, so you may want to look into that. A problem with Goto BLAS was that it would inexplicably slow down, or grind to a halt, if two separate R processes were calling the BLAS routines at the same time, whereas ATLAS never gave us any trouble. My understanding is that Goto BLAS is not maintained anymore but the existing version should work well. HTH, Peter
On 21 August 2011 at 10:18, Noah Silverman wrote: | Hi, | | I want to create a custom build of R, for my particular machine that will run as fast as possible. | | CPU: Dual cora AMD 64 bit | | OS: Fedora 15 | | I know there are a lot of options for compile time flags, lapack, blas, atlas, etc. Not sure what the best combination is. | | Has someone already worked this out? Please the fine manual at http://cran.r-project.org/doc/manuals/R-admin.html#Configuration-on-a-Unix_002dalike Dirk -- Two new Rcpp master classes for R and C++ integration scheduled for New York (Sep 24) and San Francisco (Oct 8), more details are at http://dirk.eddelbuettel.com/blog/2011/08/04#rcpp_classes_2011-09_and_2011-10 http://www.revolutionanalytics.com/products/training/public/rcpp-master-class.php