Dear Janardhan, even if you also fix the following problem (init function) you will dump into a series of problems just afterword. There is no arm of using 1060 GPUs on your HW configuration for double precision based applications such as QE.
For the sake of giving you an answer, if you have 8 cores nodes and 2 GPUS per node you can use 2 MPI per node and 4 OMP thread per process. For more information about how to execute QE-GPU in your system, please contact me personally as this topic is longer of interest of this mailing list due to the "old" HW we are considering. Kind Regards, Ivan On 25/09/2014 06:10, janardhan H.L. wrote: > Dear Spiga, > Thank you for your kind replay. > This GPU (Tesla C1060 / M1060) may be old and computation may be poor > compared to todays GPUs but still the combined speed (we have 4 nos) > could be more than multi core cpus. Even though the current QE-GPU is > not supported on this GPU > it successfully compiled but while invoking pw-gpu.x following errors > is obtained > > " ***WARNING: unbalanced configuration (1 MPI per node, 5 GPUs per node) > Error in (first zero) memory allocation , program will be > terminated!!! Bye... " > > can you please guide me to properly allocate memory between GPU and > CPU or per node (2 nodes, 8 -cores/node) > > with thanks and respectful regards > > Janardhan H L > Poornaprajna Institute of Scientific Research > Bangalore, India -------------- next part -------------- An HTML attachment was scrubbed... URL: http://pwscf.org/pipermail/pw_forum/attachments/20140925/645b0452/attachment.html
