From: JC Gumbart (gumbart_at_physics.gatech.edu)
Date: Mon Jun 27 2016 - 00:26:31 CDT
I would also go for the 3x 980s. You don’t need the higher precision or ECC for MD simulations, excepting for fatal flaws as in (2) below. One thing to note is that NAMD and Gromacs both require some CPUs too, so make sure you have at least 4-5 cores / GPU if you want to run three separate simulations simultaneously. I have a box with four GPUs and 20 cores and get much more work out of it than a single Tesla card.
Best,
JC
> On Jun 26, 2016, at 3:07 PM, Aron Broom <broomsday_at_gmail.com> wrote:
>
> I haven't used the newest gpus, but my experience would suggest the following:
>
> 1) the 3x 980 will get you better performance if you are running multiple different simulations. I don't know about fastest speed on a single simulation, but I suspect the performance advantage would evaporate there.
>
> 2) there is the possibility of memory issues with the GeForce series cards. If you go that route, you should consider using the GPU memory checking program from the Pande lab (google OpenMM or SimTK). I had a 580 card once that was crashing randomly mid simulation and yet the simulations worked fine on another card. The memory checker showed many memory issues. I've never had any problems with the workstation cards.
>
> 3) what about warranty? You might be able to afford the K80, but what if something happens to it? Are you also getting some kind of coverage? Because it would be much simpler to replace the 980s with whatever was a good price at the time.
>
> On Sat, Jun 25, 2016 at 12:24 PM, Nikhil Maroli <scinikhil_at_gmail.com <mailto:scinikhil_at_gmail.com>> wrote:
> Dear all,
>
> we are going to set up a new GPU workstation system with (Processor -2 x Intel® Xeon® E5-2670 v3 (2.3 GHz, 30 MB cache, 12 cores))
>
> we have two option
>
> 1. Tesla K80
>
> 2. 3 x GeForce GTX 980ti (Three cards connect with SLI Bridge and achieving good speed which equivalent to K80 )
>
>
> Which one i should select ,We have sufficient money for K80
>
> Awaiting reply from experienced members
>
> --
> Regards,
> Nikhil Maroli
>
>
>
>
> --
> Aron Broom M.Sc
> PhD Student
> Department of Chemistry
> University of Waterloo
This archive was generated by hypermail 2.1.6 : Tue Dec 27 2016 - 23:22:17 CST