Hi,
Yes I agree with Dan 'in principal'. As in one should try to avoid
overclocked GPUs. However that comes with a caveat that as a model matures
the yield on the silicon tends to get better and so higher clock bins
become possible reliably. What this means is that the factory spec clock
actually slides off the bottom of what is available. E.g. for the GTX-780
I don't think one can buy a GPU that isn't overclocked these days. It's
more that todays overclock (on silicon that has had 6 to 9 months to
'mature' in terms of fab quality) is effectively the same as the initial
factory release clocks.
I hope that makes sense. So in other words I would take the highest clock
model available, step back one bin in terms of frequencies (this is
normally the next one down in terms of price from the same manufacturer)
and go with that. So what that means for say Titan-Black which was only
just released. Right now I'd say avoid overclocked models, but in 6 to 9
months time when you get overclocked++ as it were then the overclocked
version is probably a safe bet.
The simple test though is to buy the one you want. Install it and then run
this:
https://dl.dropboxusercontent.com/u/708185/GPU_Validation_Test.tar.gz
(adjusting the shell file for the number of GPUs you have) and make sure
that all runs on the same model of GPU yield the same result (takes 5 or 6
hours to run). If they do then you have a good GPU. If it doesn't then RMA
the GPU.
All the best
Ross
On 5/14/14, 10:21 AM, "Daniel Roe" <daniel.r.roe.gmail.com> wrote:
>Hi,
>
>GPU experts may want to weigh in on this, but my opinion is that the
>extra cycles you get from the overclock are not going to be worth the
>potential problems it may introduce (higher temperature, decreased
>stability, etc). For playing games it's fine since you won't notice a
>calculation error on your screen out of the millions of calcs that are
>going on, but for running simulations that error can be costly.
>
>To sum up, in my opinion when running simulations I'll sacrifice some
>speed for stability any day.
>
>-Dan
>
>On Wed, May 14, 2014 at 3:41 AM, Thomas Evangelidis <tevang3.gmail.com>
>wrote:
>> Another short question: is it too risky buying an overclocked GPU? Do I
>> need to place it in a room with constant low temperature?
>>
>>
>> On 14 May 2014 00:21, Ross Walker <ross.rosswalker.co.uk> wrote:
>>
>>> Hi Thomas,
>>>
>>> The 'spread' on that survey should tell you mostly all you need to
>>>know.
>>> For the GeForce cards I don't there is really a huge difference. I
>>>tend to
>>> look for those that have the same cooling design as the Consumer Tesla
>>> cards - so that's why I have the eVGA ones listed on the website with
>>>the
>>> single ducked fan and metal body. But there is not a huge difference in
>>> any of them I believe. Pick one that looks sturdy with a reasonable
>>>price.
>>> You probably won't go wrong with major brands like eVGA, ASUS or PNY.
>>>
>>> All the best
>>> Ross
>>>
>>>
>>> On 5/13/14, 2:12 PM, "Thomas Evangelidis" <tevang3.gmail.com> wrote:
>>>
>>> >Hi guys,
>>> >
>>> >I got the message. GPUs are for single jobs.
>>> >
>>> >What about the manufacturer? I can find the same NVIDIA chip set with
>>> >varying prices according to the manufacturer. Is there noticeable
>>> >difference in the speed, cooling, etc? I found various statistics on
>>>the
>>> >internet, like this one:
>>> >
>>> >http://www.tomshardware.co.uk/forum/246796-33-what-manufacturer
>>> >
>>> >On the website you recommend EVGA GPUs. Would recommend against
>>>purchasing
>>> >the same chip set from a cheaper manufacturer?
>>> >
>>> >thanks,
>>> >Thomas
>>> >
>>> >
>>> >
>>> >
>>> >
>>> >On 12 May 2014 22:20, Ross Walker <ross.rosswalker.co.uk> wrote:
>>> >
>>> >> Hi Thomas,
>>> >>
>>> >>
>>> >> There is no advantage in terms of performance - it really is just
>>>the
>>> >> maximum system size you can simulate. What you mention with regards
>>>to
>>> >> REMD is the only other case where more memory helps you but it's
>>>really
>>> >> not advisable to overload the GPU with multiple replicas. Firstly
>>> >> everything has to fit in memory on the GPU plus there is all the
>>> >>overhead
>>> >> of task switching which will likely destroy performance. The GPU
>>>REMD in
>>> >> AMBER is really designed for running across multiple GPUs where
>>>there
>>> >>are
>>> >> at least 1 GPU per replica. Annoying I know but it was a simple way
>>>of
>>> >> figuring out how to make use of the monstrous GPU supercomputers
>>>people
>>> >> are building.
>>> >>
>>> >> All the best
>>> >> Ross
>>> >>
>>> >> On 5/12/14, 11:58 AM, "Thomas Evangelidis" <tevang3.gmail.com>
>>>wrote:
>>> >>
>>> >> >Dear AMBER community,
>>> >> >
>>> >> >Is there any other advantage of having a GPU with big memory apart
>>>from
>>> >> >the
>>> >> >maximum system size it can endure?
>>> >> >
>>> >> >http://ambermd.org/gpus/#supported_gpus
>>> >> >
>>> >> >I am not sure if it's worth running replica exchange on a single
>>>GPU
>>> >>if no
>>> >> >sufficient CPU resources are available. For example, if I have a
>>> >>single
>>> >> >workstation with an i7 processor (8 threads) and a GTX 680 2GB
>>>will it
>>> >>be
>>> >> >much faster running 8 replicas of Hamiltonian Replica Exchnage on
>>>the
>>> >>GPU
>>> >> >for a system of <100 K atoms? Will I be able to use more replicas
>>>if
>>> >>the
>>> >> >GPU has 4 GB of memory?
>>> >> >
>>> >> >thanks,
>>> >> >Thomas
>>> >> >
>>> >> >
>>> >> >PS: Btw, if anyone sells any good-performing GPU in good condition
>>> >>please
>>> >> >contact me privately.
>>> >> >
>>> >> >
>>> >> >--
>>> >> >
>>> >>
>>>>======================================================================
>>> >> >
>>> >> >Thomas Evangelidis
>>> >> >
>>> >> >PhD student
>>> >> >University of Athens
>>> >> >Faculty of Pharmacy
>>> >> >Department of Pharmaceutical Chemistry
>>> >> >Panepistimioupoli-Zografou
>>> >> >157 71 Athens
>>> >> >GREECE
>>> >> >
>>> >> >email: tevang.pharm.uoa.gr
>>> >> >
>>> >> > tevang3.gmail.com
>>> >> >
>>> >> >
>>> >> >website: https://sites.google.com/site/thomasevangelidishomepage/
>>> >> >_______________________________________________
>>> >> >AMBER mailing list
>>> >> >AMBER.ambermd.org
>>> >> >http://lists.ambermd.org/mailman/listinfo/amber
>>> >>
>>> >>
>>> >>
>>> >> _______________________________________________
>>> >> AMBER mailing list
>>> >> AMBER.ambermd.org
>>> >> http://lists.ambermd.org/mailman/listinfo/amber
>>> >>
>>> >
>>> >
>>> >
>>> >--
>>> >
>>> >======================================================================
>>> >
>>> >Thomas Evangelidis
>>> >
>>> >PhD student
>>> >University of Athens
>>> >Faculty of Pharmacy
>>> >Department of Pharmaceutical Chemistry
>>> >Panepistimioupoli-Zografou
>>> >157 71 Athens
>>> >GREECE
>>> >
>>> >email: tevang.pharm.uoa.gr
>>> >
>>> > tevang3.gmail.com
>>> >
>>> >
>>> >website: https://sites.google.com/site/thomasevangelidishomepage/
>>> >_______________________________________________
>>> >AMBER mailing list
>>> >AMBER.ambermd.org
>>> >http://lists.ambermd.org/mailman/listinfo/amber
>>>
>>>
>>>
>>> _______________________________________________
>>> AMBER mailing list
>>> AMBER.ambermd.org
>>> http://lists.ambermd.org/mailman/listinfo/amber
>>>
>>
>>
>>
>> --
>>
>> ======================================================================
>>
>> Thomas Evangelidis
>>
>> PhD student
>> University of Athens
>> Faculty of Pharmacy
>> Department of Pharmaceutical Chemistry
>> Panepistimioupoli-Zografou
>> 157 71 Athens
>> GREECE
>>
>> email: tevang.pharm.uoa.gr
>>
>> tevang3.gmail.com
>>
>>
>> website: https://sites.google.com/site/thomasevangelidishomepage/
>> _______________________________________________
>> AMBER mailing list
>> AMBER.ambermd.org
>> http://lists.ambermd.org/mailman/listinfo/amber
>
>
>
>--
>-------------------------
>Daniel R. Roe, PhD
>Department of Medicinal Chemistry
>University of Utah
>30 South 2000 East, Room 201
>Salt Lake City, UT 84112-5820
>http://home.chpc.utah.edu/~cheatham/
>(801) 587-9652
>(801) 585-6208 (Fax)
>
>_______________________________________________
>AMBER mailing list
>AMBER.ambermd.org
>http://lists.ambermd.org/mailman/listinfo/amber
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Wed May 14 2014 - 11:00:03 PDT
