[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: dual-core amd question

On Thu, 2007-01-18 at 11:12 -0800, Francesco Pietra wrote:
> Why 4 processors? Moreover, with such a system is that
> command optimized or should MPI used instead? In other
> words, is my arrangement "shared memory" or should be
> considered a cluster? 

4 processors because 2 x 2 = 4

> That because someone has recently warned me that
> "dual-opteron is not shared memory [I knew that of
> course].  Each cpu has its own memory, and they can
> access each other's. If your program uses MPI then
> that's the best way of having it."

Dual Opteron is definitely shared memory in the modern sense of the term
- it's actually NUMA, which means each processor has its own supply of
memory but can request memory from other processors (whereas with
'traditional' SMP, there's one big pool of memory and one pool of
processors). Massive 'SMP' systems such as SGI's Itanium-powered Altix
are NUMA.

Performance, however, is a bigger question. Is there a hit when
accessing a different processor's memory? Yes. Would MPI be faster?
Possibly, but probably not on a 2x2 machine. Opteron suffers from
diminishing returns (especially with a 2.6 kernel), so adding more cores
takes you further & further from optimal performance - meaning MPI
becomes a more interesting prospect at that point.

--Jo Shields

Reply to: