[R] Hardware Suggestions

[Prof Brian Ripley]

On Fri, 21 Jan 2005, Peter Dalgaard wrote:

> "Jon Dressel" <jdressel at surromed.com> writes:
>
>> We are currently running R under Windows 2000 on a server box
>> running with 2 1.2 GHZ Intel Pentium III Processors. We would like
>> to run this on a new computer running Linux and receive a
>> significant speed increase over our current implementation. Could
>> anyone provide some suggestions for a fast 64 BIT Intel based
>> processor computer with a recommendation for memory and
>> speed/type/number of processors. Also which version of R would
>> install "out-of-the-box" easily on this computer and what version of
>> Linux should be used? Thanks in advance for any help.
>
> (I assume "Intel" also means AMD?)
>
> People seem quite happy with dual and quad Opterons (and there are
> dual-core chips coming up soon, I hear), but you do need to do your
> homework, since there have been trouble with some chipsets/BIOSes in
> large-memory configurations, and there are not all that many people
> using the high-end stuff. Check out the archives of the x86_64 mailing
> lists for the popular Linux distributions.

Just find a good box-builder and let them take care of such details.
We have several dual Opterons as well as a 100+ processor cluster.

> Distribution-wise Fedora Core and SuSE both work nicely and R has been
> tested on both with no issues that I can think of. There's an RPM up
> for FC3, but it's not a big hassle to build from source and you need
> most of the build tools in place to install CRAN packages anyway.

Agreed. You do need recent versions of the OS (FC3 works better than FC2, 
and a lot better than the version of RHEL3 we returned for a refund: we 
also run SuSe 9.x).

People have been running 64-bit R for a long time on other hardware and I 
have run systematic tests across CRAN on one of our x86_64s.  All but a 
handful of maintainers have responded to my change suggestions and so 
almost all packages have passes their tests.

-- 
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