Ed Troy wrote:
>
> I am looking to upgrade my computer facility.
So am I.
> any multiprocessor PIV's, at least not yet. I like using a
> multi-processor unit because I usually have at least 7 or 8 programs
> running at one time, at least one of which is often doing some intensive
> number crunching. Thus, even though the number-cruncher does not run any
> faster than it would in a single processor unit, I can use the machine
> for multiple tasks without things slowing down. (I know the number
Why not get a fast single processor for the em sim, and use the
existing computer for the other tasks? I think that the fastest
Intel multiprocessor available is Xeon which is slower than P4.
There are also dual AMD boards running at 1.26 GHz.
> cruncher would run faster if it was multi-threaded, but I am not aware
> of many programs that are multi-threaded, at least not yet.) It also
> seems like the PIV motherboards use SDRAM, rather than RAMBUS, like some
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
I think that you have this backwards. Intel sells P4 boards with
RAMBUS. VIA has *just* announced the first P4 support chipset that
uses DDR SDRAM. I don't think that a motherboard using that chipset
is on the market yet. See, for example,
http://www.geek.com/news/geeknews/2001aug/chi20010815007311.htm
> of the PIII machines, and I would think RAMBUS would make a big
> difference in crunching numbers, especially in EM type problems, where
> the required memory can often grow very quickly.
There is a lot of debate about this. There is a latency issue: for
transfers of large contiguous chunks of memory it is faster, but
with small chunks of memory, the setup time makes it effectively
slower. So it depends on how your code is written. Also note
that RAMBUS memory is very expensive.
> Also, does anyone know a site that performs and publishes hard ratings,
> in terms of speed, of the various motherboard/processor combinations?
You can find these but they are either done with office/graphics/game
software and thus probably not representative of your computational
task,
or they are a collection of anecdotes with poorly described hardware.
My choice is not final yet, but I am leaning towards a single Athlon
with
a large amount of DDR-SDRAM.
Bruce
-- Bruce Veidt National Research Council bruce.veidt_at_nrc.ca Box 248 (250)493-2277 Penticton, BC V2A 6K3 fax: 493-7767 Canada -- The NEC-List mailing list <nec-list_at_gweep.ca> http://www.gweep.ca/mailman/listinfo.cgi/nec-listReceived on Fri Oct 12 2001 - 21:05:02 EDT
This archive was generated by hypermail 2.2.0 : Sat Oct 02 2010 - 00:10:41 EDT