From: Fred Salsbury (salsbufr_at_wfu.edu)
Date: Wed Aug 27 2003 - 13:24:11 CDT

Hello --

Thanks for your reply. I would be using this for visualization and some
data analysis; not for production runs, so it is not clear how useful your
benchmarks are for this purpose -- though I bookmarked them for future
reference on production machines.

John recommended as SGI/SUN workstation. I actually have mostly used
VMD on O2's. Does anyone have experience with the SGI Fuel workstation vs
a dual processor Xenon workstation or a dual processor Athalon
workstation, or even a dual Operton workstation?

thanks!

Fred

On Wed, 27 Aug 2003, Axel Kohlmeyer wrote:

>
>
> hello fred,
>
> i have to support john's opinion. even if you would expect a better
> performance from a dual-xeon machine, there is a high chance,
> that you would not get it. i have recently done a series of benchmarks
> on a selection of PC hardware (and some workstation machines) and have
> found, that the SMP performance on dual-xeon machines is horrible for
> any application that does not stay mostly in the cpu-cache (i.e. almost
> any serious application). on top of that you should keep in mind, that
> current PC hardware is so memory bandwidth starved, that you need
> low-latency, dual-channel memory to unleash the full power of a single
> CPU machine. you can look up some of my dual-cpu benchmark numbers at
> <http://www.theochem.ruhr-uni-bochum.de/~axel.kohlmeyer/cpmd-bench.html#smpovr>.
> so that you can see for yourself.
>
> hope this helps,
> axel.
>
> >>> "JS" == John Stone <johns_at_ks.uiuc.edu> writes:
>
> JS> Dear Fred,
> JS> I believe that the problems with SMP machines and NVidia boards
> JS> have been fixed since I wrote that email. In the last year I can
> JS> say that the NVidia SMP problem we used to see seem to have gone away.
> JS> I tested the Wildcat 7100 series drivers back in April/May and at
> JS> the time they did not support SMP kernels at all. I.e. the driver
> JS> won't even load. This was a documented limitation with the Wildcat
> JS> drivers I tested at the time. I suspect this is still the case, but
> JS> to make sure, you should send an email to the 3DLabs folks. Also,
> JS> the Wildcat Linux drivers are picky about what motherboards they support,
> JS> I had to get a loaner machine to try out with the Linux drivers because
> JS> none of our machines had the right motherboard chipset to work with the
> JS> drivers I tested.
>
> JS> Regarding CPUs, I personally favor Athlon-based systems as they
> JS> perform better on the codes I tend to run. Your mileage will vary.
> JS> The Wildcat Linux drivers don't support Athlon yet, so if you have
> JS> an Athlon, your only option would be to go with NVidia or ATI.
>
> JS> Of the 3 major vendors with Linux drivers, NVidia's drivers are
> JS> the most mature and full-featured. ATI's are next maturity-wise,
> JS> and the 3DLabs drivers are the youngest and have the most work
> JS> ahead of them as far as stability and completeness goes.
>
> JS> In the long run the 3DLabs hardware is definitely the way to go,
> JS> as their cards perform faster than the others, at least with VMD,
> JS> I can't give them much of a recommendation yet since I only got to run
> JS> test drivers, and they don't support the Athlons I have in my office.
> JS> If you're interested in the 3DLabs cards, I'd definitely suggest to get a
> JS> "demo board" to try out with the Linux drivers before you commit to
> JS> them since they weren't quite ready for running VMD last I checked.
>
> JS> For what it might cost to purchase a Xeon based machine, you could
> JS> probably afford to buy a real Unix box from Sun or SGI, with a good
> JS> graphics board even, so I'd recommend against going with Xeons.
>
> JS> I'm biased as I personally would MUCH rather use a stable machine like
> JS> a Sun/SGI box than a PC for graphics work. So, unless you've got a good
> JS> reason to blow cash on a Xeon, I'd stick with more cost-effective
> JS> options or else just buy a real workstation.
>
> JS> I hope that's somewhat useful information for you.
>
> JS> Thanks,
> JS> John Stone
> JS> vmd_at_ks.uiuc.edu
>
> JS> On Tue, Aug 26, 2003 at 08:49:41PM -0400, Fred Salsbury wrote:
> >>
> >> Hello all --
> >>
> >> I've been using VMD for a few years, but I am new to this list.
> >> I am looking to acquire a workstation primarily to use with VMD and I was
> >> looking through the archives of the mailing list, and I came across the
> >> following remark by John Stone (in reply to someone mentioning wanting to
> >> run on a dual Xenon workstation:
> >>
> >> >>>>
> >> Several of the vendors have had problems with their OpenGL drivers
> >> on SMP machines, so any comments you have about the stability you get on
> >> such machines would be very interesting.
> >> >>>>
> >>
> >> Can someone comment on this? I was pondering either one of the new dual
> >> 3.06MHz (1MB L3 Cache) Xenon workstations or a P4 3.2GHz desktop, and with
> >> either a Nvidia QuadroFX 1000 graphics card or a 3Dlabs Wildcat4 7110
> >> graphics card. So I'm wondering if there have been noticeable stability
> >> problems, and is an AMD athlon workstation would be better then.
> >>
> >> thanks!
> >>
> >> Fred
> >>
> >>
> >> Dr. Fred Salsbury
> >> Assistant Professor
> >> Department of Physics
> >> Wake Forest University
>
> JS> --
> JS> NIH Resource for Macromolecular Modeling and Bioinformatics
> JS> Beckman Institute for Advanced Science and Technology
> JS> University of Illinois, 405 N. Mathews Ave, Urbana, IL 61801
> JS> Email: johns_at_ks.uiuc.edu Phone: 217-244-3349
> JS> WWW: http://www.ks.uiuc.edu/~johns/ Fax: 217-244-6078
>
>
> --
>
> =======================================================================
> Axel Kohlmeyer e-mail: axel.kohlmeyer_at_theochem.ruhr-uni-bochum.de
> Lehrstuhl fuer Theoretische Chemie Phone: ++49 (0)234/32-26673
> Ruhr-Universitaet Bochum - NC 03/53 Fax: ++49 (0)234/32-14045
> D-44780 Bochum http://www.theochem.ruhr-uni-bochum.de
> =======================================================================
> If you make something idiot-proof, the universe creates a better idiot.
>