From: Axel Kohlmeyer (akohlmey_at_gmail.com)
Date: Thu Dec 12 2013 - 09:23:37 CST
On Thu, Dec 12, 2013 at 4:10 PM, Aron Broom <broomsday_at_gmail.com> wrote:
> I've not used an AMD GPU for any calculations, but I have tried OpenCL
> instead of CUDA with the OpenMM MD library, and in that case the performance
> is comparable. So, at least the language is not limiting as it once was.
yes, OpenCL and CUDA are not that different. in fact, LAMMPS uses an
abstraction that allows to compile the same code for either CUDA or
OpenCL with just a minimal set of special case code.
http://users.nccs.gov/~wb8/geryon/geryon.pdf
> Would certainly be interesting to see how the performance compares, I think
> the architectures are fairly different.
well, not as much as a Xeon Phi (which should also be programmable
through OpenCL at some point, IIRC). the actual problems to map
scientific applications effectively to those devices is not *that*
different. in all cases, you need to "find" huge amounts of
concurrency. that and the complexity of the computation would
determine how well you can hide latencies and memory bandwidth
differences.
the most attractive aspect of the whole development to me is that
accelerators make particularly "expensive" force fields "cheaper". in
other words, complex calculations can be accelerated much better than
streamlined force fields like charmm, amber and alike.
axel.
>
> On Thu, Dec 12, 2013 at 2:05 AM, Norman Geist
> <norman.geist_at_uni-greifswald.de> wrote:
>>
>> Jerome,
>>
>> as far as I know, without rewriting the NAMD code, you can't use it with
>> AMD(ATI) GPUs. NAMD is written in CUDA which belongs to Nvidia. AMD cards
>> won't run CUDA code. A language both types of GPUs work with is openCl. But
>> CUDA is faster than openCl on Nvidia GPUs. So there can't barely be a person
>> with experience in running namd on GPUs other than Nvidia.
>>
>> Norman Geist.
>>
>> > -----Ursprüngliche Nachricht-----
>> > Von: owner-vmd-l_at_ks.uiuc.edu [mailto:owner-vmd-l_at_ks.uiuc.edu] Im
>> > Auftrag von Jérôme Hénin
>> > Gesendet: Mittwoch, 11. Dezember 2013 18:31
>> > An: vmd
>> > Betreff: vmd-l: AMD vs. nvidia support
>> >
>> > Hi everyone,
>> >
>> > Traditionally I used to favor nvidia GPUs over AMD for running VMD
>> > under Linux, because the drivers were "better" (that is, they ran with
>> > some glitches, but ran nonetheless). The landscape seems to be evolving
>> > these days, and nvidia has notoriously antagonized the Linus crowd (or
>> > was that Linux?) for its lack of cooperation, while AMD has been more
>> > forthcoming. That was mostly a problem with open source support, but I
>> > wonder if that also reflects in the quality of the proprietary drivers.
>> >
>> > So, it would be great to hear about the recent experiences of some
>> > VMD/Linux users with AMD and nvidia support - both in terms of features
>> > and performance.
>> >
>> > Thanks,
>> > Jerome
>>
>>
>> ---
>> Diese E-Mail ist frei von Viren und Malware, denn der avast! Antivirus
>> Schutz ist aktiv.
>> http://www.avast.com
>>
>>
>
>
>
> --
> Aron Broom M.Sc
> PhD Student
> Department of Chemistry
> University of Waterloo
-- Dr. Axel Kohlmeyer akohlmey_at_gmail.com http://goo.gl/1wk0 International Centre for Theoretical Physics, Trieste. Italy.
This archive was generated by hypermail 2.1.6 : Wed Dec 31 2014 - 23:22:00 CST