Tyan's Tiger MPX motherboard : review - link
Robert G. Brown
rgb at phy.duke.edu
Thu Mar 14 07:53:07 PST 2002
On Wed, 13 Mar 2002, Donald B. Kinghorn wrote:
> On Wednesday 13 March 2002 14:09, Robert G. Brown wrote:
> I think Bob Drzyzgula has most of the answers in his comments about 3.3v vs
> 5v operation of the slots ... I hadn't even thought about that ...
> The system I just brought up for testing is 10 nodes + a master using the
> Tyan 2466's. I put it in 3U rack cases so I wasn't constrained by risers ...
> I disabled the onboard 3com nic and put an intel eepro100 on each of the
> nodes but put them in 32/33 slots ... I had a bunch of SIS 6326 video cards
> AGP and PCI but the board wouldn't post with either of these cards installed.
> I used an ATI rage PCI card in a 64/66 slot to bring the systems up to flash
> the bios to the 1.01 version and then put the SIS cards in and haven't had
> any trouble since.
> I'm getting ready to do a system in 2U cases so I'll have to deal with the
> riser card problems soon. It will probably be a week before I get a chance to
> do anymore playing around with the boards so hopefully someone else can get
> some answers to you sooner.
> My gut feeling is that there is a problem with the MSI motherboard though.
> Have you had a chance to look at any of the overclocker sites to see if
> anyone has had a chance to beat the board up yet? Some of those guys are
> pretty thorough :-)
Well, I actually have to encumber my grant ASAP (use it or lose it
money) and plan to cut a PR today for 16 dual nodes. Since you say to
DID get the Tiger systems to come up with an ATI Rage in a 64/66 slot,
it adds to the evidence that the problem is with the MSI and not the
basic chipset or the ATI card itself, although I'll still do some more
tests to try to find out.
Here is my overall CBA comparing the two:
MSI MS-6501 (760 MPX/762)
a) One of two boards on AMD's "approved dual/762MP mobo's" web page.
The other is an ASUS, not the Tiger -- don't ask me why. Actually, I've
ASKED our local AMD rep and hope to hear soon -- one would like to think
that there is some sort of an objective test suite used to determine
presence on this page instead of corporate back-scratching or paid
advertising, and if so I'd be very interested to know just what test the
b) Appears to have trouble with at least one 32/33 vidoe card (ATI
Rage) on the 64/66 bus, probably associated with a failure to readjust
its voltage from 3.3 to 5v. Result: unpredictable failure.
c) No serial console option in the bios.
d) Available with "optional" onboard NIC (eepro100 from the look of
it), but our local vendor doesn't have any in this configuration,
pricewatch doesn't offer any in this configuration, and we haven't been
able to test it (for free, at any rate:-). Curious thing to make an
option on a server class board -- seems like its presence would make it
a DIFFERENT motherboard. At any rate, LAN option may be vapor at this
e) Pricewatch minimum price about $215.
f) At least one satisfied customer on the beowulf list (IIRC).
In my opinion, given c) the lack of reliable video on a 64/66 is a
killer b) -- a show stopper for a 1U or 2U configuration. If the 64/66
slots didn't matter, one would get the cheaper Tiger 2460 or equivalent.
Tiger 2466 (also 760MPX/762):
a) Not "approved dual motherboard" by AMD for reasons unknown. Does it
fail some test?
b) Does NOT fail to install with ATI Rage in 64/66 slot, apparently
c) Serial console option in the bios, so don't need video card anyway
(probably). This saves the equivalent of at least $30 in the overall
price AND means that doesn't have to mix card widths and speeds on the
2 64/66 slots on a 2U riser, which in turn might well force the whole
bus to run at the slower speed.
d) Standard onboard 3Com NIC. (Question to list: Do we need to
reflash this to get PXE to work? I ask because empirically all the
3c905 CARDS we are getting come with v4.0 which will NOT PXE BOOT, and
have to be reflashed to 4.30 in order to PXE boot/install. If we have
to reflash, I will need to get a floppy, which otherwise I would prefer
e) Pricewatch minimum $218, but by the time one removes the cost of a
3c905 100BT card and video card from a configuration (required for MSI
to "match" its functionality) it is more like $140 (or to compare apples
to apples, the MSI card's real "price" in application is more likely to
be around $280-290)
f) Several happy customers on beowulf list, at least one VERY happy
customer (JBLP) right here at Duke.
On the basis of this, using a Tiger instead of an MSI is a no-brainer
UNLESS there is some hideous failure revealed by AMD's testing. I would
think this unlikely for two reasons: a) People on this list like the
MoBo, and are likely to hammer it hard and in ways that are similar to
how I will hammer it. If it fails some obscure test in a configuration
that I'll never encounter, who cares? b) This is Tyan's second try at
this and by now they should have gotten it right. The 2460 was
definitely flawed from many reports, although ours have worked "all
right" so far. The 2466 SHOULD repair most of those flaws -- even
recently made 2460's seem to work OK these days.
So unless I hear differently, I'll be ordering Tyans. Thanks to all who
responded and participated in this thread over the last 3 weeks.
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