[R] ATLAS for Pentium D and Pentium Duo Core?

Uwe Ligges ligges at statistik.uni-dortmund.de
Thu Mar 8 08:26:59 CET 2007



Jonathan Zhang wrote:
> Dear all,
> 
>   I wanted to use ATLAS to increase the computational speed of my
> program. I have installed R in a straightfoward way with .exe file
> (not building from source).
> 
>   However, after getting the P4 ATLAS Rblas.dll file from the CRAN
> directory bin/windows/contrib/ATLAS, and replacing the default
> Rblas.dll file in my R home directory under bin/, the speed actually
> decreased by 3% compared with the default Rblas file!!


Such a 3% thing might happen even for the right BLAS given you are not 
doing heavy vector/matrix operations.


>   So what am I doing wrong? Is the above the correct procedure for
> using ATLAS? Or, is my machine not a Pentium 4 and hence the ATLAS
> file that I got from CRAN is not appropriate?
> 
>   If the issue lies in the processor, then please kindly let me know
> where to get the correct ATLAS file for the following computers:
> 
>  IThe first one is Pentium D 3.2 ghz, the second is a MacBook 1.83ghz
> dual-boot and I am running R in windows.

The MacBook is running Windows? So it is a newer one containing some 
Intel Core2Duo CPU?

For the Pentium D, the P4 ATLAS should be more or less appropriate. Of 
course, you can compile one that fits even better to your system.
For the MacBook, we do not have an appropriate BLAS on CRAN. What you 
could try is the Pentium M BLAS (which should come close), but we cannot 
guarantee stability nor performance.
We are looking for volunteers to compile a BLAS for each architecture, 
so please contribute!

Uwe Ligges


>   Thank you!
> 
> Jonathan Zhang
> 
> Marketing Division
> Columbia Business School
> 
> ______________________________________________
> R-help at stat.math.ethz.ch mailing list
> https://stat.ethz.ch/mailman/listinfo/r-help
> PLEASE do read the posting guide http://www.R-project.org/posting-guide.html
> and provide commented, minimal, self-contained, reproducible code.



More information about the R-help mailing list