First of all thanks to Peter Dalgaard, Thomas Lumley and Prof Brian Ripley
for quick responses.
Indeed, I will be running Windows2000 on this machine (my company's
standard). I will also install the full version of cygwin on it.
When I asked my original question I thought about just copying the P4 dll
from the ATLAS directory on CRAN. As I understand from the three answers I
got, this should work, but perhaps not be the fastest BLAS possible. When
I get the machine (beginning of April) I will try to configure and compile
ATLAS and Goto and compare the results.
Thanks again,
Andy
__________________________________
Andy Jaworski
518-1-01
Process Laboratory
3M Corporate Research Laboratory
-----
E-mail: apjaworski at mmm.com
Tel: (651) 733-6092
Fax: (651) 736-3122
|---------+----------------------------->
| | Prof Brian Ripley |
| | <ripley at stats.ox.a|
| | c.uk> |
| | |
| | 03/12/2004 16:16 |
| | |
|---------+----------------------------->
>-----------------------------------------------------------------------------------------------------------------------------|
|
|
| To: Peter Dalgaard <p.dalgaard at biostat.ku.dk>
|
| cc: apjaworski at mmm.com
|
| <R-help at stat.math.ethz.ch>
|
| Subject: Re: [R] Xeon CPU and ATLAS
|
>-----------------------------------------------------------------------------------------------------------------------------|
AFAIK all P4s are SSE2.
He wants a Windows ATLAS build, and that is not hard, but you have to do
that under Cygwin not MinGW.
We have a Xeon, but have found Athlons and now Opterons better value.
I don't have access to Windows Xeon machine: anyone who does could
contribute a Xeon version of Rblas.dll (to Uwe Ligges, please).
Note that for 1.9.0 you can use Goto's BLAS and he as a P4 version that
is faster than ATLAS (under Windows). So I would start with that.
Although Thomas Lumley is right about cache sizes, using a P4 512kb ATLAS
or Goto BLAS is pretty good on a Xeon (only tested by me under Linux).
Using a BLAS for too large a cache is disastrous, but underestimating by a
factor of 2 is not bad at all. (I think most Xeons are 1Mb cache, with
Xeon MP being 4Mb.)
On 12 Mar 2004, Peter Dalgaard wrote:
> apjaworski at mmm.com writes:
>
> > I am about to get a new machine at work - an IBM Intellistation with
the> > Xeon 2.8 GB processor. It will run Windows 2000. I would like to
install> > the proper ATLAS dll for this machine, but I am not sure if Xeon is
P4?
> > Does anybody have any experience with Xeon?
>
> Yes, we have a dual 2.8 Xeon running Linux. Very nice machine. We've
> had it for a bit more than a year and it's been of very good use. I
> can see that I did two ATLAS builds in feb 2003, one is called
> Linux_P4SSE2_2 and the other is Linux_Xeon_2. AFAIR (but it's been a
> while), the former is the preconfigured thing that came out of a
> quick-install and the other one is fully tuned version, and there was
> no difference worth speaking of between the two.
>
> I think the difference between P4 and Xeon is the SSE1 vs SSE2 thing,
> so you'd want the P4SSE2 DLL, but others may be able to speak more
> authoritatively.
>
>
--
Brian D. Ripley, ripley at stats.ox.ac.uk
Professor of Applied Statistics, http://www.stats.ox.ac.uk/~ripley/
University of Oxford, Tel: +44 1865 272861 (self)
1 South Parks Road, +44 1865 272866 (PA)
Oxford OX1 3TG, UK Fax: +44 1865 272595