Cabinets for rackmount cases

Robert G. Brown rgb at phy.duke.edu
Sat Mar 16 10:26:29 PST 2002


On Thu, 14 Mar 2002, Velocet wrote:

> of systems in real cases? in 4,2 and 1U cases? How much extra is getting
> all AMD-aproved gear? Doesnt this limit us to not buying Tyan Tiger gear
> which is unapproved?

Curiously, I just had a chat with our local AMD rep and they do indeed
have a test suite that must be passed for "approval".  Apparently the
Tiger is in the queue, but the rep couldn't tell me in detail whether it
was failing any tests at all let alone which ones it might be failing as
"AMD works with the vendor" to move the systems into compliance with
spec but AMD obviously doesn't want to shoot any of its primary mobo
vendors in the foot during the testing period.

FWIW, after discovering the problem with at least one 32/33 card in a
64/66 slot on the MSI board we experimented with moving cards around on
our Tiger 2460's, and learned that even there it makes a difference
which cards are in which slots even with these.  This was very
interesting because there has been a huge range of experience reported
on this list with 2460's -- some people swear that they are completely
trouble free and work perfectly and stably and others report that they
run and crash, run and crash.

Poor (non-registered-ECC) memory and the use of XPs instead of MPs have
been proposed as possible causes, but there are still those that do fine
with XPs and others who do fine with at least one or two sticks of OTS
memory.  Perhaps some or all of the stability problems were caused by
having cards in "the wrong slots" (not that there should BE any wrong
slots, but maybe there are anyway and if so this should be empirically
accepted and dealt with...).

This is hardly unknown in the history of OTC PCs and PCI implementations
-- my old Supermicro dual PPros would simply not work with my Tulip NICs
in certain PCI slots.  Sure, it was impossible according to PCI spec,
but the systems would not install or would install and then crash in use
for "no reason" if they were in certain slots and work perfectly (for
years) if they were in other slots.  Naturally, I used the slots that
worked.

   rgb

-- 
Robert G. Brown	                       http://www.phy.duke.edu/~rgb/
Duke University Dept. of Physics, Box 90305
Durham, N.C. 27708-0305
Phone: 1-919-660-2567  Fax: 919-660-2525     email:rgb at phy.duke.edu






More information about the Beowulf mailing list