When you run in single precision, do you get -6.77 for E_pair on step 0?
Can you run a memory test to make sure that there is not a hardware issue? Please see:
http://lammps.sandia.gov/threads/msg28376.html
This lj melt benchmark has been run on a lot of hardware and I don’t see an obvious problem here. Thanks. - Mike
Hi Chen,
did you run a memory test on your card?
If you checked out the discussion thread:
http://lammps.sandia.gov/threads/msg28376.html
you would see where to download the cuda memtest:
http://sourceforge.net/projects/cudagpumemtest/
and what memtest reports if there's any hardware issue with the card.
To verify if the output from running the example scripts are
consistent with what supposed to be, you can compare them with those
in the log files available in the corresponding folders: for example,
with the script in.melt, there are two log files under examples/melt.
Cheers,
-Trung