[CP2K-user] A rare CUDA bug in 6.1

[Anton Kudelin]

Hi all!

Since some moment I have getting a program fault (CUDA Error: invalid 
argument) in some cases when ELPA has been setting as a diagonalization 
library. The bug seems to appear when system size exceeds a certain threshold, 
for instance, if one ethanol molecule would be removed from pos.xyz (see 
attachment) calculation will start normally. Another condition is the 
number of ranks/threads, in my case the calculation breaks when np>4 and 
OMP_NUM_THREADS>2 (except 5x3). Does anybody know why CP2K 6.1 behave such 
a way?

Anton K.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20190110/c55bd0e3/attachment.htm>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: DEBUG-5-4
Type: application/octet-stream
Size: 456349 bytes
Desc: not available
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20190110/c55bd0e3/attachment.obj>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: DEBUG-6-3
Type: application/octet-stream
Size: 457134 bytes
Desc: not available
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20190110/c55bd0e3/attachment-0001.obj>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: INP
Type: chemical/x-gamess-input
Size: 1178 bytes
Desc: not available
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20190110/c55bd0e3/attachment.inp>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: pos.xyz
Type: chemical/x-xyz
Size: 7502 bytes
Desc: not available
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20190110/c55bd0e3/attachment.xyz>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: VCT.inc
Type: application/octet-stream
Size: 135 bytes
Desc: not available
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20190110/c55bd0e3/attachment-0002.obj>