[R] xeon processor and ATLAS

[Douglas Bates]

On 8/29/07, hui xie <sealights30 at yahoo.com> wrote:
> Thanks very much for all your advice. To be clear, my OS is window XP. I bought this server last year. It's Dell Precision PWS690. THe processor is Xeon(TM) CPU 3GHZ, 2G RAM. I am not sure how to check more details of processor on my computer. But I went to Dell website and from what I can recall, it seemed that I ordered a Dual-core Intel Xeon 5160 3GHz, 1333FSB, 4MB L2 Cache, 80watts.

> I think from time to time, I will need linear algebra to do things such as Choleskey Factorization, and matrix inverse etc... At the same time, I considered myself quite unskilled on building R by myself. It seems there were lots of details that I can get things wrong. So if there is an existing ATLAS one on R website that I can use, I would be very happy to use it to replace the default one. The reason why I would like to use ATLAS is that R FAQ said :

Rather than worrying about versions of ATLAS you may want to spend
your time considering exactly how you are going about the
calculations.  A general result in numerical linear algebra is that if
your algorithm involves computing the inverse of a matrix you have a
sub-optimal algorithm.

> "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 guess my  Xeon processor is either a P4 or  Core2Duo, but I am really not sure which one to use. Could you please offer me some suggestions?
>
> Again, many thanks for all your advice!
>
> Best,
>
> Hui
>
> Uwe Ligges <ligges at statistik.uni-dortmund.de> wrote:
>
> Prof Brian Ripley wrote:
> > On Tue, 28 Aug 2007, hui xie wrote:
> >
> >> hi everyone:
> >>
> >> I have a Dell Server that has a Xeon processor, and I would like to use
> >> the best ATLAS posted in the R website. I find that R has ATLAS for
> >> core2duo and P4. I am not sure which one of these two is best suited for
> >> Xeon processor, or is that neither of these two is good and I should
> >> stick with the default one that was installed originally?
> >
> > And your OS is?
> >
> > There are many different 'Xeon' processors with very different
> > capabilities.
>
> ... the earlier similar to P4 and some similar to Core2Duo. You won't
> make use of bigger L2/L3 caches in Xeon processors.
>
>  > You really ought to build ATLAS for yourself if numerical
> > linear algebra performance matters to you (and it makes little difference
> > to most people: I think Uwe Ligges quoted 10% for testing all CRAN
> > packages).
>
>
> Right, it depends on what you are really doing. If most time is spend in
> certain numerical matrix operations, ATLAS is your friend. In all other
> cases, it does not matter so much, as Brian cited correctly.
>
> Uwe Ligges
>
>
> >
> >> Your advice is very much appreciated!
> >>
> >> Best,
> >>
> >> Hui
> >>
> >>
> >> ---------------------------------
> >> Park yourself in front of a world of choices in alternative vehicles.
> >>
> >>  [[alternative HTML version deleted]]
> >>
> >> ______________________________________________
> >> 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.
> >
> > Please do!
> >
> >
>
>
>
> ---------------------------------
>
>         [[alternative HTML version deleted]]
>
> ______________________________________________
> 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.
>