[LA-SiGMA-gpu] Fwd: RE: Fwd: namd-l: Nvidia GPUs

Wed Aug 21 11:43:06 CDT 2013

Hi,

Here comes the reply from Mark at NVIDIA. His email was bounced due
to the size of the attachment. If you have any questions, please send
email to Mark directly (see his email address below).
Mark, Jon, and Jonathan from NVIDIA are not subscribed to this list.

The slides can be found at
https://dl.dropboxusercontent.com/u/11746493/LA-SiGMA/NVIDIA_Applications.pdf
(7.2MB)

Regards,
Jian

-------- Original Message --------
Subject: 	RE: [LA-SiGMA-gpu] Fwd: namd-l: Nvidia GPUs
Date: 	Wed, 21 Aug 2013 09:16:02 -0700
From: 	Mark Berger <mberger at nvidia.com>
To: 	Jon Saposhnik <jsaposhnik at nvidia.com>, Jian Tao <jtao at cct.lsu.edu>,
"Thomas C. Bishop" <bishop at latech.edu>
CC: 	Jonathan Bentz <jbentz at nvidia.com>, "lasigma-gpu at loni.org"
<lasigma-gpu at loni.org>

We are thankful for your interest in and use of GPUs to run NAMD (and
other applications).

You might also consider Tesla GPUs, which while more expensive, are
built to run 24 X 7 in clusters.  GeForce boards are designed, built and
tested to run the 18+ hours a day that a gamer is awake.  If you are
building a large cluster with many nodes or planning to run long
simulations then you should consider the benefits of Tesla GPUs.  In
addition, NVIDIA, CRAY and the Blue Waters team at UIUC have put a lot
of effort into strong scaling on Tesla GPUs on the Blue Waters and Titan
systems.

Jon Saposhnik has supplied you with slides that express NVIDIA’s
position on GeForce vs. Tesla for clusters.

Please take a look at the attached slides which contain lots of
information about computational chemistry applications on Tesla.

Best regards,
Mark

*From:*Jon Saposhnik
*Sent:* Wednesday, August 21, 2013 9:01 AM
*To:* Jian Tao; Thomas C. Bishop
*Cc:* Jonathan Bentz; lasigma-gpu at loni.org; Mark Berger
*Subject:* Re: [LA-SiGMA-gpu] Fwd: namd-l: Nvidia GPUs

Hi All,

Adding in Mark Berger from NVIDIA .  Mark works with the Computational
Chemistry ISV's.

Mark can answer NAMD questions.

As to Tesla versus GTX:  Tesla is Server GPU product line w. enhanced
set of Hardware and Software functionality features plus additional
support.

This deck has some comparison for GTX Titan to Tesla K20/K20X.

https://docs.google.com/file/d/0B2MSwnI8sm0nWWdLc0t4N1FscDQ/edit?usp=sharing

Tier 1 HPC Server vendors  presently don't qualify nor support GTX Cards
in their Servers , though you may not care. (HP, DELL, IBM, SGI, CRAY)

I'd think one would want to run their own reliability and benchmark
suites over long periods of time should they care to scientificaly know
if GTX (Titan or 700 series) cards are acceptable for their Heterogenous
cluster use.

Regards,

Jon Saposhnik

NVIDIA Industry Business Development

Higher Education and Research

Jsaposhnik at NVIDIA.com <mailto:Jsaposhnik at NVIDIA.com>

949.212.6491 Cell

*GTC 2014 | March 24-27 | San Jose, California*

Sign up for news and announcements –www.gputechconf.com/emaillist
<http://www.gputechconf.com/emaillist>

View the growing list of GPU-accelerated applications at
www.nvidia.com/teslaapps
<applewebdata://A809A530-7524-4C34-962A-F9433EFAD263/www.nvidia.com/teslaapps>

*From: *Jian Tao <jtao at cct.lsu.edu <mailto:jtao at cct.lsu.edu>>
*Date: *Tuesday, August 20, 2013 8:17 PM
*To: *"Thomas C. Bishop" <bishop at latech.edu <mailto:bishop at latech.edu>>
*Cc: *Jon Saposhnik <jsaposhnik at nvidia.com
<mailto:jsaposhnik at nvidia.com>>, Jonathan Bentz <jbentz at nvidia.com
<mailto:jbentz at nvidia.com>>, "lasigma-gpu at loni.org
<mailto:lasigma-gpu at loni.org>" <lasigma-gpu at loni.org
<mailto:lasigma-gpu at loni.org>>
*Subject: *Re: [LA-SiGMA-gpu] Fwd: namd-l: Nvidia GPUs

Hi Tom,

A GTX cluster could be very cost-effective for some applications.

I heard from a collaborator from the Poznan supercomputing and
networking center (PSNC) in Poland that they did build a GTX cluster a
couple years ago. I could check to see how things are working.

Some of us at CCT are collaborating with researchers from the Charm++
group in other projects. We know that their team is helping the NAMD
group to port and optimize NAMD on GPUs. There have been a lot of work
on GPUs lately though a lot of their own efforts are on high-end GPU
cards mainly because they are actively involved in the Blue Waters project.

As you may have heard, LSU HPC will very soon make available a half
million dollar Kepler cluster, Shelob. The LA-SiGMA project is one of
the main science drivers of the Shelob proposal. You and other LA-SiGMA
teams who are doing GPU related research shall have access to the system.

IMHO, effectively sharing such resources is economically better than
hosting a personal cluster if cost is one of the major considerations.
While the LSU GPU team might not help you much on NAMD itself, we will
work together with LSU HPC to at least help your team to get started on
the system once it is ready.

Regards,

Jian

-- From my phone (no network at home today)

On Aug 20, 2013, at 2:32 PM, "Thomas C. Bishop" <bishop at latech.edu
<mailto:bishop at latech.edu>> wrote:

      Jon and Jonathan and LASIGMA-gpu

      There is great interest the  NAMD user community in using
      gtx grade cards for personal clusters instead of Kepler/Tesla cards.
      (see namd-l forward)

      It would be of real interest to molec. modelers to have
      benchmarks w/ diff cuda cards using NAMD's standard benchmark systems.
      and specifically look for instances of HW failure , esp. memory
      failures, comparing gtx to server grade cards.

      This would be something I could pursue w/ the LASiGMA/LSu GPU group.

      Tom

      -------- Original Message -------

      *Subject: *

      namd-l: Nvidia GPUs

      *Date: *

      Tue, 20 Aug 2013 14:27:28 -0400

      *From: *

      Thomas Albers <talbers at binghamton.edu> <mailto:talbers at binghamton.edu>

      *To: *

      Namd Mailing List <namd-l at ks.uiuc.edu> <mailto:namd-l at ks.uiuc.edu>

      Dear List,

      we are in the early stages of planning for a new cluster.  Would

      anyone on this list have experience (and timing results) to share with

      recent Nvidia GPUs (GTX 660Ti, GTX 680, GTX 7xx)?

      Regards,

      Thomas

      _______________________________________________
      LASiGMA-gpu mailing list
      LASiGMA-gpu at loni.org <mailto:LASiGMA-gpu at loni.org>
      https://mail.loni.org/mailman/listinfo/lasigma-gpu
      wiki: https://wiki.cct.lsu.edu/la-sigma/
      http://www.institute.loni.org/lasigma/lasigma-gpu.php

------------------------------------------------------------------------
This email message is for the sole use of the intended recipient(s) and
may contain confidential information.  Any unauthorized review, use,
disclosure or distribution is prohibited.  If you are not the intended
recipient, please contact the sender by reply email and destroy all
copies of the original message.
------------------------------------------------------------------------