Wanted: Good mobo for Intel 850E chipset and 1066 RDRAM

Jeff Nguyen jeff at aslab.com
Fri Aug 30 12:18:18 PDT 2002


You should get P4T533-C instead of P4T533. The latter requires
RIMM4200 modules which are harder to get and cost more. In
addition, there are only 2 RIMM slots in P4T533. That would
require you to purchase higher cost, higher density memory to
support 1GB/2GB.

Since these boards are designed for the low cost, desktop market,
there won't be any support for 64-bit slots. Maybe that will be
change in the coming years.

Jeff

ASL Inc.


----- Original Message -----
From: "Camm Maguire" <camm at enhanced.com>
To: <josip at icase.edu>
Cc: <beowulf at beowulf.org>
Sent: Friday, August 30, 2002 11:52 AM
Subject: Re: Wanted: Good mobo for Intel 850E chipset and 1066 RDRAM


> Greetings, all, and many thanks for your helpful replies.
>
> OK, we're thinking about the Asus P4T533 motherboards with 1066 RAMBUS
> memory.  This is a 16-node cluster upgrade.  Our old network was
> dual-channel-bonded 100Mbit switched ethernet.  In general, we leave
> the network unsaturated by a fair margin on our critical code, but do
> find that the network is a slight bottleneck on small problems.
>
> We're considering a gigabit network upgrade, probably using the Intel
> PRO/1000T adapters together with the driver in the latest 2.4.20-pre
> kernel.  Unfortunately, the Asus motherboards only have 32bit/33Mhz
> PCI slots.  As best as we can figure, this will shave a factor of 2
> off of the peak possible network throughput otherwise attainable.
> Which means that we could expect a rough factor of 2 remaining in
> throughput over our existing setup, with no improvement in latency.
>
> The nodes are going from PIII 850Mhz with SDRAM to P4 2.53 and
> RAMBUS.  As far as we can tell, this amounts to a factor of 6 for cpu
> intensive work (taking into account the extra factor of 2 provided by
> the SSE2 instructions), and something less but still high on
> memory-bottleneck problems.  So the capability of the machine is being
> tilted further in the node capacity direction relative to the network
> capacity.
>
> We're trying to decide if we should take a 30% hit on memory speed in
> exchange for a factor of 2 in network throughput by going to RAMBUS
> PC800 motherboards with a faster PCI slot.  Of course, if anyone knows
> of a motherboard supporting both RAMBUS 1066 and either 64bit PCI or
> 66Mhz PCI, that would be ideal, but such animals apparently do not yet
> exist.
>
> Any advice, suggestions, or commentary are most appreciated.
>
> Take care,
>
> P.S.  We're looking at the Netgear GS516T gigabit switch.  So far
> we've been unable to discern whether this switch supports 'jumbo
> frames' or whether such settings are even necessary with the Intel
> cards.  Any illumination here would be appreciated as well!
>
>
> Josip Loncaric <josip at icase.edu> writes:
>
> > I forgot to mention one more thing: high speed memory like PC1066 tends
to run
> > hot.  Users suggest to have good airflow over the memory modules (e.g.,
keep
> > wiring out of the way).
> >
> > Sincerely,
> > Josip
> >
> > --
> > Dr. Josip Loncaric, Research Fellow               mailto:josip at icase.edu
> > ICASE, Mail Stop 132C           PGP key at http://www.icase.edu./~josip/
> > NASA Langley Research Center             mailto:j.loncaric at larc.nasa.gov
> > Hampton, VA 23681-2199, USA    Tel. +1 757 864-2192  Fax +1 757 864-6134
> >
> >
>
> --
> Camm Maguire      camm at enhanced.com
> ==========================================================================
> "The earth is but one country, and mankind its citizens."  --  Baha'u'llah
> _______________________________________________
> Beowulf mailing list, Beowulf at beowulf.org
> To change your subscription (digest mode or unsubscribe) visit
http://www.beowulf.org/mailman/listinfo/beowulf
>




More information about the Beowulf mailing list