[gmx-users] GPU vs Processor load

[Pappu Kumar]

I want to run one simulation in GTX 780Ti utilizing six threads of overclocked Haswell-E 5820. So I will be running two simulations in my computer using two 780Ti. The system size varies between 60k to 300k atoms. My question is if six threads will be good enough to keep the balance between CPU and GPU load. Or I should go for a slower GPU like GTX 780 to keep the balance. Thanks.

On Wed, Aug 13, 2014 at 1:55 PM, Pappu Kumar <papuu_k at yahoo.com> wrote:
>Thanks for your info. I am planning to buy one upcoming Haswell E-5820K to >use with 2x GTX 780Ti if feasible. So there will be six threads/GPU. I am >wondering if it will be good enough or I should go for 2x GTX 780.  Did you mean the other way around, the 780Ti is considerable faster
than the 780? What kind of systems/simulations will you be running? If you plan to
use the machine for single runs and/or relatively small sysmtes, a
single 780 Ti may be better. That's because at the moment in v4.6/5.0
you need domain-decomposition to use >=2 GPUs and due to the initial
domain-decomposition overhead (1 to 2 domains), the scaling from one
to two GPUs is OK, but not great (and typically the smaller the
simulation system the worse). We are working on making it possible to use multiple GPUs within a
node without domain-decomposition for the next version. >I have >plans to overclock the CPUs by 10-20%.  That should work quite well. --
Szilárd >>On Tuesday, 12 August 2014 9:11 PM, Szilárd Páll <pall.szilard at gmail.com> >wrote: >>>Yes, it definitely will. The difference will be quite pronounced in >CPU-only runs, but if you use a GPU this will shrink and could become >relatively small depending on the turbo/frequency scaling settings >you'll use. >-- >Szilárd >>>On Tue, Aug 12, 2014 at 9:37 AM, Pappu Kumar <papuu_k at yahoo.com> wrote: >>I am planning to buy an Intel Haswell-E 5820 processor which supports >>AVX2. I am wondering if it will have better performance than i7 4930K which >>supports AVX. Thank you.