[HPGMG Forum] Execution heterogeneity
jed at jedbrown.org
Wed Apr 30 18:39:34 UTC 2014
Brian Van Straalen <bvstraalen at lbl.gov> writes:
> That is a whole other growing issue, algorithm heterogeneity. I’m thinking architecture.
> Latency cores vs throughput cores, partitioned local memory, co-processing.
I'm not entirely sure how to cleanly support this. Coarse levels should
use latency-optimized cores while the finest levels can use
throughput-optimized cores. A common address space would simplify this.
I'm not too concerned about MIC in offload mode; I think it's enough to
focus on native. We definitely need CUDA in the short-to-medium-term,
though programming model aside, I have doubts about the attainable HPGMG
performance on that hardware.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Size: 835 bytes
Desc: not available
More information about the HPGMG-Forum