> You should not guide your infrastructure investments on SP/DP FLOPs/clk/core
> metric.

for sure!  especially because these peak FP rates are for cache-friendly,
optimally-vectorizable codes.  I'm not arguing that no one does large 
matrix-matrix operations, for instance, but rather that a lot of codes 
have hot loops that don't vectorize - or at least not 8 to 16-wide...

regards, mark hahn.
_______________________________________________
Beowulf mailing list, [email protected] sponsored by Penguin Computing
To change your subscription (digest mode or unsubscribe) visit 
http://www.beowulf.org/mailman/listinfo/beowulf

Reply via email to