[R] xeon processor and ATLAS
Jeffrey J. Hallman
jhallman at frb.gov
Thu Aug 30 15:33:32 CEST 2007
hui xie <sealights30 at yahoo.com> writes:
> The reason why I would like to use ATLAS is that R FAQ said :
> "The savings can be appreciable: on a 2.6GHz P4 and a 1000 x 1000 matrix svd took 16.2 sec with the standard BLAS and 7.8 sec with ATLAS. Because ATLAS is tuned to a particular chip we can't use it generally: the optimal routines for a P4 or an Athlon XP are quite different and neither will run at all on a PII."
> This seems to me an impressive gain to use the correct ATLAS instead of the
> default BLAS.
I've been doing econometrics for nearly 20 years, and have not yet run across
a situation that called for looking at a 1000 x 1000 matrix. I tend not to
believe analyses with more than a dozen explanatory variables. But I suppose
that you could run across a situation where a design matrix with hundreds of
columns actually made sense. But does that really happen often enough that
saving 8 seconds of computation time is worth going through the ATLAS hassle?
More information about the R-help