On Fri, 23 Jun 2006 07:02:52 -0300
"Vinicius Santos" <[EMAIL PROTECTED]> wrote:
> > We thought about this, but the OGA architecture has just way too many
> > things trying to talk to one block of memory controllers. And we
> > don't even way to TRY to connect two chips to the same set of RAMs.
> What about the AMD way? ie. each cpu having direct access to separated
> RAM banks and doing some sort of transaction making it basically a
> NUMA architecture wich act externally as SMP. (I may be wrong about
> that, but that's what I got from reading some specs of Athlon X2)
> I'm just wondering the implications, not even suggesting as a real
> implementation idea. :)
All multi processor systems are inherently complicated.
No matter whether it's SMP, NUMA or a cluster
(read a good book on distributed systems, like
"Distributed Systems - Principles and Paradigms" from Tannenbaum
and you'll know why).
IMHO the effort and difficulty to make a multi processor
system work is too difficult for OGP. At least at this stage.
Let us first build a simple and fool proof system that _works_
and let us after that think how we can get more power out of it.
Attila Kinali
--
心をこめて聞け心をこめて話せ
_______________________________________________
Open-graphics mailing list
[email protected]
http://lists.duskglow.com/mailman/listinfo/open-graphics
List service provided by Duskglow Consulting, LLC (www.duskglow.com)
