VMD-L Mailing List

From: John Stone (johns_at_ks.uiuc.edu)
Date: Wed Apr 06 2005 - 16:01:34 CDT

Hi,
  One thing to be aware of is that PCI-X (not express, just "X") is
a _significantly_ slower interconnect than AGP or PCI-Express. If you're
using several 3-D video cards at once in such a machine, unless each of
the video cards is in a slot that's part of a _separate_ PCI-X bus, you'll
completely swamp the bus with graphics I/O which will bring the machine
to its knees. High-end workstation machines like Sun and SGI boxes
typically use several separate PCI busses, so this issue is resolved on
the big machines, but you'll have to look VERY carefully at the design
of PC machines to verify that they would be usable for multi-headed
PCI-X-based graphics displays. The situation should improve dramatically
if/when we start seeing new machines with multiple PCI-Express slots for
graphics boards. There are already a few such motherboards, currently
tailored towards the NVidia multi-board "SLI" configuration, those are
what I'd probably keep an eye on in the short-term. Longer term I expect
the workstation vendors to offer big multi-headed machines based on
Opterons etc.

Regarding CPU choice, I definitely recommend the new 64-bit CPUs for VMD....
There's NO reason not to buy a 64-bit processor at this point. If you're
worried about backwards compatibility with software, you can run a 32-bit OS
until you're comfortable going full 64-bit. Our entire research lab now
consists of a mix of 64-bit Sun Opterons running Linux and Sun Sparc machines
running Solaris, which are the machines we do our visualization on.

For other programs like BLAST or NAMD, the choice is less clear.
In the case of NAMD, the main interest is speed. NAMD doesn't use
much memory for normal simulations, so 64-bit CPUs aren't as big of an
advantage as they are for VMD. For BLAST I assume 64-bit would be an
advantage due to the increased word size and resultant increase in
pattern matching speed per clock cycle, and max process RAM use,
but you'd need to run the 64-bit version of course.

  John Stone
  vmd_at_ks.uiuc.edu

On Wed, Apr 06, 2005 at 03:39:19PM -0500, Sabuj Pattanayek wrote:
> Hi,
>
> Karsten Suhre wrote:
> > Ideally, I would be looking for a 4-processor board, but it seems there
> > is no such with AGP or PCI-Express support (correct?).
>
> You can get 4 cpu opteron (pogolinux, tyan, arima, hpc systems)
> racks/boards or 8 cpu opteron racks (pogolinux, arima, armari, hpc
> systems, etc) all with pci-x slots (64 bit pci this is not pci-express).
>
> The pci-x slots can be used with *multiple* nvidia quadro 600 pci cards
> which have stereo support (since pci-x slots are backwards compatible
> with pci). You can compare its specviewperf pro-e/ugs/3dsmax
> specifications / performance with other nvidia quadro video cards on
> their website.
>
> I've never seen or heard of anyone using such a configuration for a
> visualization machine but it doesn't mean it can't be done.
>
> > How about the processor - is a 64-bit Opteron a good choice? Or would
> > you stick to 32 bit for compatibility reasons?
>
> You can always use a 32 bit chroot for linux applications that are
> distributed in binary only format and that don't run in a 64-bit
> environment.
>
> Hope that helps,
> Sabuj Pattanayek 

-- 
NIH Resource for Macromolecular Modeling and Bioinformatics
Beckman Institute for Advanced Science and Technology
University of Illinois, 405 N. Mathews Ave, Urbana, IL 61801
Email: johns_at_ks.uiuc.edu                 Phone: 217-244-3349              
  WWW: http://www.ks.uiuc.edu/~johns/      Fax: 217-244-6078