Gigabyte GA-8KNXP Ultra-64 not so Ultra

hrudy

Distinguished
Aug 19, 2004
17
0
18,510
The Gigabyte GA-8KNXP Ultra-64 has a bottleneck. Namely the Hublink 1.5, which connects the NorthBridge to the Hance Rapids 6300ESB Southbridge, runs at 266 MB/SEC.

This link distributes bandwidth among the PCI-X 64/66 bus, PCI 32/33 bus and onboard peripherals such as the SCSI controller. The bottom line is that the PCI-X plug in cards won’t run anywhere near its rated 528 MB/SEC. We found this out the hard way by trying to use HDTV frame buffers in the PCI-X slots. No wonder that Intel never made a motherboard with this chipset combination.
Another triumph of marketing over engineering.
 
I found out empirically by running a PCI-X Bus Speed test using a PCI 64/66MHZ HDTV frame buffer plug in board. The results of the test were 200 MB/sec much slower than on a 7505 dual Xeon motherboard. This is despite the fact that the 875 supports much faster memory speeds.

The Hublink issue was identified by a motherboard rep and later confirmed by an Intel rep.
 
I own this board, just bought 6-weeks ago. So far, it is the best board I have owned, although the most expensive. I'll look into this just so I can understand the impact. Would there be any way to overclock it up to speed?

BTW, Toms article on this board, page 3, stated a con that it did not include SCSI cable within package. On that same page the pic shown shows that cable, and I got one with my board.
The Link to article...
http://www.tomshardware.com/firstlook/20040817/index.html
<P ID="edit"><FONT SIZE=-1><EM>Edited by RichPLS on 08/23/04 06:05 PM.</EM></FONT></P>
 
I was considering buying this board but the post about the bottleneck has is making me think a little longer about it. If it only affects the PCI-X cards then its no big deal for me. Anyone know if that bottleneck would affect SCSI performance (In a RAID 0 situation)?

And BTW, Gigabyte's website lists the SCSI cable as part of the package.
 
This is the fastest and feature laden m/b I have seen!!! I do not know for sure, but I suspect that this bottleneck is a mistaken implication.
If someone knows more, please post links and backup.
Rich