Re: [AMBER] GTX-Titan Black Edition Timings

From: filip fratev <filipfratev.yahoo.com>
Date: Fri, 21 Feb 2014 14:19:41 -0800 (PST)

Hi Ross,

I have no words...Amazing numbers for Amber 14. I have no patience to upgrade!


Regards,
Filip
 




On Friday, February 21, 2014 11:21 PM, Ross Walker <ross.rosswalker.co.uk> wrote:
 
Hi All,

For those of you who might be interested in the performance of the new
NVIDIA GTX-Titan Black edition the following are the numbers I get for
AMBER 12 on the following system with 4 GTX-Titan Blacks (PCI-E Gen 3 and
2 x E5-2650 v2
) in it. http://exxactcorp.com/index.php/solution/solu_detail/121


JAC_PRODUCTION_NVE - 23,558 atoms PME
-------------------------------------

CPU code 16 cores: |        ns/day =      12.65  seconds/ns =    6828.99
  [0] 1 x GTX-Titan-Black: |        ns/day =    118.58  seconds/ns =
  728.63
  [1] 1 x GTX-Titan-Black: |        ns/day =    119.19  seconds/ns =
  724.90
  [2] 1 x GTX-Titan-Black: |        ns/day =    121.95  seconds/ns =
  708.51
  [3] 1 x GTX-Titan-Black: |        ns/day =    120.59  seconds/ns =
  716.46
      2 x GTX-Titan-Black: |        ns/day =    125.51  seconds/ns =
  688.38
      3 x GTX-Titan-Black: |        ns/day =    139.82  seconds/ns =
  617.96
      4 x GTX-Titan-Black: |        ns/day =    163.57  seconds/ns =
  528.21

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

CPU code 16 cores: |        ns/day =      3.27  seconds/ns =  26461.08
      1 x GTX-Titan-Black: |        ns/day =      34.07  seconds/ns =
2536.31
      2 x GTX-Titan-Black: |        ns/day =      38.31  seconds/ns =
2255.47
      3 x GTX-Titan-Black: |        ns/day =      44.10  seconds/ns =
1959.02
      4 x GTX-Titan-Black: |        ns/day =      52.56  seconds/ns =
1643.98

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

CPU code 16 cores: |            ns/day =      0.67  seconds/ns =
129055.61
      1 x GTX-Titan-Black: |            ns/day =      7.90  seconds/ns
=  10932.92
      2 x GTX-Titan-Black: |        ns/day =      9.24  seconds/ns =
9351.84
      3 x GTX-Titan-Black: |        ns/day =      9.19  seconds/ns =
9400.96
      4 x GTX-Titan-Black: |        ns/day =      11.66  seconds/ns =
7410.58

NUCLEOSOME_PRODUCTION - 25,095 atoms GB
---------------------------------------

CPU code 16 cores: |            ns/day =      0.05  seconds/ns =
1789186.93
      1 x GTX-Titan-Black: |        ns/day =      3.80  seconds/ns =
22737.20
      2 x GTX-Titan-Black: |        ns/day =      4.98  seconds/ns =
17347.62
      3 x GTX-Titan-Black: |        ns/day =      6.53  seconds/ns =
13224.34
      4 x GTX-Titan-Black: |        ns/day =      8.10  seconds/ns =
10662.91

Not bad. Note however, that 1 of the 4 cards out of the box failed my
certification (and had to be swapped out) so if you plan on buying your
own from newegg, Amazon etc I'd advise testing them by running multiple
long runs (I use JAC NPT for 1,000,000 steps) with the same random seed
and repeating it about 20 times upon receipt. You should get 20 sets of
identical answers. If you don't the card is wonky and should be RMA'd (or
used for a pure gaming system). We've seen this before with bleeding edge
new GeForce cards so it's not too alarming. We can just filter the bad
ones and in my experience over time the number of cards that misbehave out
of the box rapidly shrinks to 0%. The good news at least is that this new
card does not seem to introduce any serious issues such as what we had
with the original launch of the GTX-Titan. :-)


And for those that want a little 'taster' - here are some provisional
numbers from the AMBER 14 development tree on the same hardware. (note
current limit for the new peer to peer approach we use for PME on multiple
GPUs is 2 on current hardware so you can run 4x1 GPU, 2x1+1x2GPU or 2x2
GPU runs at the same time on a node.)

JAC_PRODUCTION_NVE - 23,558 atoms PME
-------------------------------------

CPU code 16 cores: |        ns/day =      12.74  seconds/ns =    6782.04
  [0] 1 x GTX-Titan-Black: |        ns/day =    142.99  seconds/ns =
  604.25
  [1] 1 x GTX-Titan-Black: |        ns/day =    143.36  seconds/ns =
  602.68
  [2] 1 x GTX-Titan-Black: |        ns/day =    144.44  seconds/ns =
  598.16
  [3] 1 x GTX-Titan-Black: |        ns/day =    142.72  seconds/ns =
  605.39
[0,1] 2 x GTX-Titan-Black: |            ns/day =    212.38  seconds/ns
=    406.82
[2,3] 2 x GTX-Titan-Black: |            ns/day =    213.38  seconds/ns
=    404.91

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

CPU code 16 cores: |        ns/day =      3.26  seconds/ns =  26507.19
      1 x GTX-Titan-Black: |        ns/day =      40.81  seconds/ns =
2117.03
      2 x GTX-Titan-Black: |        ns/day =      61.28  seconds/ns =
1409.84

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

CPU code 16 cores: |            ns/day =      0.68  seconds/ns =
126458.12
      1 x GTX-Titan-Black: |        ns/day =      9.37  seconds/ns =
9216.86
      2 x GTX-Titan-Black: |        ns/day =      13.82  seconds/ns =
6251.74

NUCLEOSOME_PRODUCTION - 25,095 atoms GB
---------------------------------------

CPU code 16 cores: |            ns/day =      0.05  seconds/ns =
1786425.96
      1 x GTX-Titan-Black: |        ns/day =      3.84  seconds/ns =
22478.47
      2 x GTX-Titan-Black: |        ns/day =      7.45  seconds/ns =
11596.74
      3 x GTX-Titan-Black: |        ns/day =      10.06  seconds/ns =
8589.49
      4 x GTX-Titan-Black: |        ns/day =      12.35  seconds/ns =
6995.90


Throw in the Hydrogen Mass Repartitioning support and 4fs timestep and one
will be at 400ns/day+ on 2 cards.

All the best
Ross


/\
\/
|\oss Walker

---------------------------------------------------------
|            Associate Research Professor              |
|            San Diego Supercomputer Center            |
|            Adjunct Associate Professor              |
|        Dept. of Chemistry and Biochemistry          |
|          University of California San Diego          |
|                    NVIDIA Fellow                    |
| http://www.rosswalker.co.uk | http://www.wmd-lab.org |
| Tel: +1 858 822 0854 | EMail:- ross.rosswalker.co.uk  |
---------------------------------------------------------

Note: Electronic Mail is not secure, has no guarantee of delivery, may not
be read every day, and should not be used for urgent or sensitive issues.
<<<Disclaimer: Exxact contributes in part to funding my research>>>






_______________________________________________
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 Fri Feb 21 2014 - 14:30:02 PST
Custom Search