Re: [AMBER] Replica Exchange on a single Titan-X GPU

From: Ross Walker <ross.rosswalker.co.uk>
Date: Mon, 23 Mar 2015 11:28:30 -0700

We are still working on the optimization (lots to validate) but here's a sneak preview of the Titan-X performance so far for those interested.

JAC_PRODUCTION_NVE - 23,558 atoms PME 4fs
-----------------------------------------

      [0] 1 x GPU: | ns/day = 426.14 seconds/ns = 202.75
 P2P 2 x GPU: | ns/day = 525.17 seconds/ns = 164.52

JAC_PRODUCTION_NPT - 23,558 atoms PME 4fs
-----------------------------------------

      [0] 1 x GPU: | ns/day = 381.35 seconds/ns = 226.56
 P2P 2 x GPU: | ns/day = 495.13 seconds/ns = 174.50

JAC_PRODUCTION_NVE - 23,558 atoms PME 2fs
-----------------------------------------

      [0] 1 x GPU: | ns/day = 206.89 seconds/ns = 417.61
 P2P 2 x GPU: | ns/day = 277.40 seconds/ns = 311.46

JAC_PRODUCTION_NPT - 23,558 atoms PME 2fs
-----------------------------------------

      [0] 1 x GPU: | ns/day = 193.72 seconds/ns = 446.01
 P2P 2 x GPU: | ns/day = 255.99 seconds/ns = 337.52

FACTOR_IX_PRODUCTION_NVE - 90,906 atoms PME
-------------------------------------------

      [0] 1 x GPU: | ns/day = 62.92 seconds/ns = 1373.08
 P2P 2 x GPU: | ns/day = 81.92 seconds/ns = 1054.65

FACTOR_IX_PRODUCTION_NPT - 90,906 atoms PME
-------------------------------------------

      [0] 1 x GPU: | ns/day = 60.35 seconds/ns = 1431.74
 P2P 2 x GPU: | ns/day = 75.90 seconds/ns = 1138.30

CELLULOSE_PRODUCTION_NVE - 408,609 atoms PME
--------------------------------------------

      [0] 1 x GPU: | ns/day = 14.53 seconds/ns = 5946.91
 P2P 2 x GPU: | ns/day = 19.44 seconds/ns = 4444.03

CELLULOSE_PRODUCTION_NPT - 408,609 atoms PME
--------------------------------------------

      [0] 1 x GPU: | ns/day = 13.94 seconds/ns = 6199.30
 P2P 2 x GPU: | ns/day = 18.21 seconds/ns = 4745.04

All the best
Ross


> On Mar 23, 2015, at 8:35 AM, filip fratev <filipfratev.yahoo.com> wrote:
>
> Hi,Look here:http://ambermd.org/gpus/benchmarks.htm#Benchmarks
> As you can see the Titan X shows, at least for me, disappointing results -about 23% more than the "original" Titan. Ross can comment what one can expect after the Maxwell optimization but even after that GTX980 looks better solution if you don't want to waste money. To best of my knowledge the multi GPU system is the better option for Replica exchange (one GPU for each replica).
> Regards,
> Filip
>
>
> On Monday, March 23, 2015 5:19 PM, amr hamed <schrodinger8383.gmail.com> wrote:
>
>
> Dear all,
> I would like to know the performance of running a replica exchange
> simulation on a single gpu (namely the latest nvidia gpu, the titan-x which
> has 12 GB of ram).
> What do you recommend for this type of simulation: one titan-x or 2 gtx 980?
>
>
>
> Best Regards
> _______________________________________________
> 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


_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Mon Mar 23 2015 - 11:30:03 PDT
Custom Search