AMDZone web-site sends word that Advanced Micro Devices company recently rolled out C0 revision of its AMD-762 North-Bridge and also issued a new Revision Guide for AMD-762 System Controller where indicated that there is an erratum with B0 and B1 revisions of the same chip.

Apparently, there may be failures with certain add-on cards when the chipset transacts PCI or AGP memory operations that are strongly ordered between two processors in different order than intended by software because of high bus latency caused by heavy bus traffic. The failure only occurs when two CPUs are installed and running, but the issue was only identified by AMD itself, as the source reports here.

No idea if the problem has affected any commercial products or emerged in a lot of the systems, but AMD has fixed it. Maybe since it was fixed about two years after the initial release of AMD-762, it was so minor that nearly no end-users have experienced it?