On Thu, 20 Dec 2007, Martijn van Oosterhout wrote:
> The way around this is a NUMA architecture, but that's a whole
> other ball of wax.
Quick note for those reading Ulrich's paper: he refers in a couple of
places to Intel's upcoming CSI approach to NUMA. This has now been
renamed QuickPath, and it looks like it will be late 2008 before that even
makes it to Itanium processors.
The fact that AMD has a good NUMA implementation in their Opteron lines
while Intel's Xeon processors do not is one area AMD still has a clear
competative lead on. But you need memory bandwidth starved application
before that matters more than the fact that the current Xeons are faster
in general.
--
* Greg Smith gsmith@gregsmith.com http://www.gregsmith.com Baltimore, MD
