Hi,
In making your decision you need to take into account how the cpu or cpus
communication to everything else.
If you are using a Quad Core cpu then all four cores are accessing the same
shared memory, so applications which generate lots of memory accesses will slow
down the cpu as cores have to wait for their own data. However applications
which are cpu bound and have minimal access to memory will run at the full
speed of the cores. [Note: Intel profiles many hundreds of applications to
ensure that those instructions run quickly { told by a developer once },
similarly Intel's compilers are tuned to produce code which use the quickest
instructions on their cpus. Never heard if AMD perform a similar profiling
analysis].
If you are using two Dual Cores which share a common memory bus this is similar
to the Quad Core above but slower due to detecting simultaneous bus access off
chip.
If you are using two Dual Cores which do not share a common memory bus like an
AMD Opteron or Intels Core i7 cpus then each cpu has its own memory bus (no
contention between cpus), however each cpu can access the other cpus memory via
the HypherTransport (AMD) or QuickPath (Intel) connection. So applications
which are in the local memory to cpu will run more quickly but take a hit
accessing another cpus memory.
You need to understand what applications need to be run and how they operate so
you can make an informed choice.
--
This message posted from opensolaris.org
