[AMBER] AMBER12 test discrepancy

From: Guanglei Cui <amber.mail.archive.gmail.com>
Date: Tue, 3 Dec 2013 10:55:44 -0500

Hi all,

I recently compiled AMBER12 (with the latest updates) with Intel Compilers
(ver. 14.0.0), OpenMPI (ver. 1.6.5) and Cuda SDK (5.0.35). While doing the
test, I noticed some potential failures for the MPI (3 out of 459) and Cuda
(8 out of 89) runs. I've attached the diff and log files for both. Some of
the discrepancies are in the 4th decimal place, but others are greater,
especially for the CUDA code (SPFP). I wonder if anyone could kindly let me
know if this is the outcome one typically expects or I should try to
resolve this. Any help is much appreciated!

Best regards,

AMBER mailing list

Received on Tue Dec 03 2013 - 08:00:03 PST
Custom Search