adda icon indicating copy to clipboard operation
adda copied to clipboard

Published 20 hours ago • verified publisher icon adda-team

Parallel OpenCL

Open GoogleCodeExporter opened this issue 9 years ago • 3 comments 

The next big step in OpenCL mode is to make it capable of employing multiple 
GPUs. The GPUs can be either on a single machine - then this can be done using 
OpenCL alone, or in a cluster - then MPI is probably required.

Large clusters of GPUs are becoming more and more popular - so this is a way to 
high-throughput simulations.

Original issue reported on code.google.com by yurkin on 14 Jan 2014 at 9:45

GoogleCodeExporter avatar Aug 12 '15 07:08 GoogleCodeExporter

Original comment by yurkin on 3 Aug 2014 at 5:02

  • Added labels: Component-Logic

GoogleCodeExporter avatar Aug 12 '15 07:08 GoogleCodeExporter

/cc @DaveOri

myurkin avatar Jul 10 '18 06:07 myurkin

A related, but to some extent opposite idea, is to use only part of the GPU cores for simulations (controlled by a new command line option). This may allow running several ADDA instances (with relatively small memory requirements) in parallel on a single GPU (more efficiently than when each process tries to occupy the whole GPU).

myurkin avatar May 03 '23 13:05 myurkin