[Beowulf] Supermicro 2U

Nifty Tom Mitchell niftyompi at niftyegg.com
Mon Feb 23 15:58:57 EST 2009


On Mon, Feb 23, 2009 at 08:30:24AM -0800, Lucas Sheehan wrote:
> On Sun, Feb 22, 2009 at 11:53 PM, Nifty Tom Mitchell
> <niftyompi at niftyegg.com> wrote:
> >
> > In general a landed on motherboard chip set solution can be less expensive
> > than a MB+card solution.   The lack of a PCIe connector and riser can
> > improve PCIe signal quality.   The IB chip set is close to the 4x connector
> > which is a good thing.
> >
> 
> Is there really that much effect from a riser or other connector?  The
> physical connection, bus length and frequency in use seems to hint to
> me that there would be little if any loss or degradation in the
> signal.  However I've never investigated such things in depth.  Have
> you or anyone else seen performance hits because of such hardware
> configuration?

Less of an issue for performance than reliability.

It is possible to reset and recover a PCIe bus and continue.
If this happens very often there is a big performance hit.
Vias and connector contacs all subtract from the maximum PCIe 
trace length of about 20 inches and add complexity in the matching
of the apparent lengths.   These bit rates are hard.

Vias for IB are so problematic that all the chip vendors I know
have an electrical bit to flip the driver pair so the PWB guy does
not have to flip them with an unnecessary trip through a via
to and from another layer.

If bit error rates go up performance does degrade with resends.  If it gets too
bad then the device can go off line.   Better is just better...


-- 
	T o m  M i t c h e l l 
	Found me a new hat, now what?

_______________________________________________
Beowulf mailing list, Beowulf at beowulf.org
To change your subscription (digest mode or unsubscribe) visit http://www.beowulf.org/mailman/listinfo/beowulf

-- 
This message has been scanned for viruses and
dangerous content by MailScanner, and is
believed to be clean.



More information about the Beowulf mailing list