Hi Parker,
These differences are perfectly normal - if you look it is all within the
last one or two decimal places (slightly more for nucleosome but that's a
VERY large system) and only on a few steps in the various runs (as opposed
to every step). If you want to fully convince yourself it is not an issue
try running the tests on another 780 card if you have one. You should see
the exact same number of possible failures and the differences should be
identical.
The differences come from different order of operation caused by different
hardware and or different cuda / driver / compiler versions etc.
Thus you should be good to go.
All the best
Ross
On 1/30/14, 3:36 PM, "de Waal, Parker" <Parker.DeWaal.vai.org> wrote:
>Hello Everyone,
>
>I recently purchased a GTX 780 for one of our newer workstations and am
>concerned that there may be something wrong with the card.
>
>After running the test_amber_cuda.sh script 89 file comparisons passed
>while 8 file comparisons failed (diff file located here:
>https://gist.github.com/anonymous/fc3df9cd7a006556ace5).
>
>Has anyone experienced this before? Additionally do these differences
>mean that I will be unable to reliably use this card for simulation work?
>
>Cheers,
>Parker
>_______________________________________________
>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 Thu Jan 30 2014 - 16:00:03 PST
