Re: [AMBER] GTX 780 Ti

From: Ross Walker <ross.rosswalker.co.uk>
Date: Sat, 23 Nov 2013 14:14:54 -0800

Hi Filip,

The timings are only accurate for runs longer than 60 seconds. Ideally
benchmark runs should be between 2.5 and 5 minutes to get reliable
timings, especially on the new GPUs that adjust the clock speed based on
temperature - those tend to start out fast and then throttle back unless
you have decent cooling. Can you adjust the value of nstlim so it runs for
at least 2.5 mins and try again for trpcage and myoglobin please. The same
is true for DHFR and FactorIX if those took less than 60 seconds.

All the best
Ross



On 11/22/13 8:50 PM, "filip fratev" <filipfratev.yahoo.com> wrote:

>Hi Ross and Scott,
>
>
>>Nice! - Now just wait for the next version of AMBER running on two of
>>those cards with peer to peer. :-)
>
>I really wait for the new version ! :) Hope that there will be a great
>GPU scale! :)
>
>>If you could possibly run the GB tests as well - TRPCage, Myoglobin and
>>Nucleosome it would be really helpful - then I can put them on the
>>website.
>
>Ok, I have never tested GB and observed a significant difference between
>the numbers in mdinfo and output files for the first two tests. I provide
>info for both of them (mdinfo is the second one):
>
>
>TRPCage = 833.11/ 745.41 ns/day
>
>Myoglobin = 231.62 /222.63 ns/day
>
>
>Nucleosome =3.99 / 3.99 ns /day
>
>
> output file:
>| Final Performance Info:
>| -----------------------------------------------------
>| Average timings for last 249000 steps:
>| Elapsed(s) = 51.62 Per Step(ms) = 0.21
>| ns/day = 833.48 seconds/ns = 103.66
>|
>| Average timings for all steps:
>| Elapsed(s) = 51.85 Per Step(ms) = 0.21
>| ns/day = 833.11 seconds/ns = 103.71
>| -----------------------------------------------------
>
>mdinfo:
>
>--------------------------------------------------------------------------
>----
>| Current Timing Info
>| -------------------
>| Total steps : 250000 | Completed : 1000 | Remaining : 249000
>|
>| Average timings for last 1000 steps:
>| Elapsed(s) = 0.23 Per Step(ms) = 0.23
>| ns/day = 745.41 seconds/ns = 115.91
>|
>| Average timings for all steps:
>| Elapsed(s) = 0.23 Per Step(ms) = 0.23
>| ns/day = 745.41 seconds/ns = 115.91
>|
>|
>| Estimated time remaining: 57.7 seconds.
>
>--------------------------------------------------------------------------
>----
>
>
>
>
>On Friday, November 22, 2013 7:56 PM, Ross Walker <ross.rosswalker.co.uk>
>wrote:
>
>Nice! - Now just wait for the next version of AMBER running on two of
>those cards with peer to peer. :-)
>
>
>If you could possibly run the GB tests as well - TRPCage, Myoglobin and
>Nucleosome it would be really helpful - then I can put them on the
>website.
>
>Thanks for the info.
>
>All the best
>Ross
>
>On 11/22/13 12:03 AM, "filip fratev" <filipfratev.yahoo.com> wrote:
>
>>Hi all,
>>Some benchmarks of EVGA GTX 780 Ti SC. PCI2.0, CPU 2600K, Cuda 5.0:
>>
>>DHFR NVE = 126.02 ns/day
>>DHFR NPT = 97.24 ns/day
>>
>>Factor IX NVE =35.76 ns/day
>>Factor IX NPT =28.62 ns/day
>>
>>Cellulose NVE = 8.43 ns/day
>>Cellulose NPT = 6.85 ns/day
>>
>>Regards,
>>Filip
>>_______________________________________________
>>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



_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Sat Nov 23 2013 - 14:30:02 PST
Custom Search