TR Forums

I was digging through the errata on the 760 MP and MPX northbridge when I found this entry.

48 AGP Transactions are Non-functional

Products Affected. B0

Description. AGP cycles are non-functional in this silicon revision due to a silicon development tools issue.

Potential Effect in System. The system may hang when attempting to use AGP cards.

Suggested Workaround. AGP cards can still be used in PCI mode (including PCI mastering) but actual AGP cycles do not. To use AGP cards, do not load the AGP miniport driver in the system. This will force the AGP card to run only PCI cycles.

Resolution Status. This will be fixed in a future silicon revision.

The bold emphasis is mine.

A. Do you know what stepping of the 762 Northbridge you have?
B. Did you install the AMD AGP miniport driver?

If you have the B0 instead of B1 stepping and you installed the miniport driver, I would suppose that, that combination could be creating the NMI error.


_________________
<a href="http://www.geocities.com/ryu_connor/systems.htm">Ryu's PCs</a>

<font size=-1>[ This Message was edited by: Ryu Connor on 2002-01-05 17:12 ]</font>

Thats a pretty serious bug to let out the door. It would piss me off to no end to have a nice new mobo and a top of the line GFX card only to find out it was barely useable.

On 2002-01-05 17:32, SecretSquirrel wrote:
Thats a pretty serious bug to let out the door. It would piss me off to no end to have a nice new mobo and a top of the line GFX card only to find out it was barely useable.

Well, before anyone jumps to conclusions. I can't say for sure that the B0 stepping of the 762 Northbridge ever shipped retail. The chipset only has two steppings: B0 and B1. The former has this issue, the latter does not. Seeing as the 762 is a dual processor board aimed at professionals who use rather expensive AGP cards for CAD/CAM and the like, it is very possible that B1 was the first shipping revision and B0 just went to manufactures for validation.

We don't have all the details; yet.

After thinking some more about this you are correct. There are two catagories of people who these boards would be targeted at. First you have the power users: CAD/CAM and the like. For their sake (and AMD's) I hope that stepping never made it out the door. The other catagory is made up of people who are going to use the chips in server class machines. For them, graphics generally don't matter anyway.

Given the fact that the first I have heard about this bug is your posting, which you got from reading the errata, I sorta doubt that many of the buggy stepping made it out.

Based on this blurb over at the Inquirer, http://www.theinquirer.net/05010203.htm, looks like there are more than just AGP problems. Doesn't say what rev of the chipset though.

B0 was beta/eval only. The AGP bug was pretty fierce but non-lethal. It killed AGP, but the system kept running. Yellow exclamation next to the AGP controller in Windows and all that.

B1 is the shipping 762. No known AGP issues, and the NMI has been tracked back to the graphics drivers acting very naughty. It's been seen with Nvidia, ATI, and Kyro, on everything from a 440ZX up to an i860 Tyan board.

It is certainly not AMD's fault in any way. They've checked, and they're conforming to the AGP spec 100% in the areas in question, while Nvidia and friends are tossing the rule book and fighting dirty. There's a lot of discussion going on at AMD right now about whether it'd be better to hack in a workaround or just force Nvidia to fly right. The NMI is a different symptom of the same problem that brings you most of your inf. loop bugs. Those are AGP related, too.

SecretSquirrel - That's 760MPX only. 760MP has plenty of documented errata, but the most serious one a production board might encounter relates to single word DMA (rarer than hell anymore). If you're thinking about going 760MP now that MPX is dropping prices, it'd probably be a good deal.