On Wed, Jun 08, 2011 at 12:07:25AM +0800, Bojun Huang wrote:
> Thank you very much for your explainations, it's very interesting to know the
> logarithmic growth and linear loss for SMP systems (which implies that the
> benefit from adding more cores would quickly disappear in large-scale systems)
But you also need to realize what the constants are. From 128 threads
with shared memory up, you may start hitting obstacles like this, but
for ordinary machines up to 24 threads, scaling can be almost perfect.
And many-thread machines are extremely rare (especially if they are
oriented for high computation performance, i.e. not IA64); I think
BlueFuego is the only program that had a go on a >100 core PPC machine
and BlueGene/P.
So, for practical purposes, scaling with shared memory can be pretty
much perfect.
--
Petr "Pasky" Baudis
UNIX is user friendly, it's just picky about who its friends are.
_______________________________________________
Computer-go mailing list
[email protected]
http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
