NVidia chipsets (was Re: Any experience with "AsusA8N-SLi"or"Gigabyte K8NXP-SLI" mobos?)

[David O'Brien]

On Wed, Feb 16, 2005 at 07:47:14AM +0100, Lars Tunkrans wrote:
> David O'Brien wrote:
> >On Tue, Feb 15, 2005 at 08:13:30PM -0600, Astrodog wrote:
> >
> >>On Wed, 16 Feb 2005 12:14:14 +1100, Jan Mikkelsen
> >><janm at transactionware.com> wrote:
> 
> >>I'm not sure anyone but AMD makes SMP-capible chipsets for Opteron.
> >
> >Note that the HP dual-processor workstation doesn't use the AMD chipset,
> >yet is MP.  (Opteron's also aren't SMP, but are a NUMA architecture)
> 
> NUMA is the opposite of UMA ( Uniform Memory Architecture)   not the 
> opposite of SMP.
> I see SMP as an oppsite of GRID-clusters like Beowulf.

You're the only one then.  Google for "smp numa".  Most in the NUMA world
treat SMP the opposite.  ccNUMA (cache-coherent) isn't "symmetric" in
its memory latency or I/O paths, thus not SMP.  FreeBSD treats MP Opteron
as SMP.  Note that a single package(socket) AMD dual-core CPU is indeed
SMP.

http://whatis.techtarget.com/definition/0,,sid9_gci212678,00.html
    An SMP system is a "tightly-coupled," "share everything" system in
    which multiple processors working under a single operating system
    access each other's memory over a common bus or "interconnect" path.

This does not define a NUMA system.  A 2nd source:

http://lse.sourceforge.net/numa/faq/
    3. What is the difference between NUMA and SMP?
    The NUMA architecture was designed to surpass the scalability limits
    of the SMP architecture. With SMP, which stands for Symmetric
    Multi-Processing, all memory access are posted to the same shared
    memory bus.

While technically correct about "UMA" (remove the 'N' in NUMA and "UMA" is
what you get).  In the AMD64 context, UMA, mean on-board graphics and the
"unified" speaks of video RAM and system RAM.

 
> These days a NUMA architecture is radically faster than a UMA architecture 
..
> Anyone, 10 years ago, saying that NUMA 
> is faster than UMA
> would have been looked at as out of his mind.

Uh, no.  You design a NUMA architecture exactly to get some very fast
memory to processors.

-- 
-- David  (obrien at FreeBSD.org)