Hello,
We are getting the above error when trying to start minimisation of a
630,000 atom structure on a Tesla M2090 GPU using Amber 14 patch 13.
The minimisation runs without error on a different system at the same
Amber patch level with a TITAN-X card. Output from pmemd.cuda is attached.
I found a thread relating to this error on older cards here:
http://archive.ambermd.org/201410/0184.html . Although there is mention
there of a patch being issued, I can't find it in the Amber 14 patch
list. The key change appears to be a change to the definition of lblocks
at around line 400 in amber14/src/pmemd/src/cuda/kPMEInterpolation.cu
(I'm finding it a little hard to discern from the thread whether other
chanegs are required):
from
int lblocks = min(blocks, 65536);
to
int lblocks = min(blocks, 65535);
On the system in error, the definition has not been changed: it is still
min(blocks, 65536).
Did a fix for this problem ever make it in to the official patches?
Many thanks
William
William Lees
Associate Research Fellow
Institute of Structural and Molecular Biology
Birkbeck, University of London
http://shepherd-group.ismb.lon.ac.uk
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Wed Aug 10 2016 - 02:30:02 PDT
