There is, however, a most beautiful K8W mobo from Tyan over at this Korean page, with many detailed shots and specifications.
It seems that when the key elements of two Opteron memory controllers, a NUMA aware OS, such as Suse64, Windows Server 2003 or Windows XP-64 beta, and an AGP slot are combined, the AGP slot is bumped out of it's normal place as PCI device 0 (zero). This creates a scenario where most video card drivers do not know where the AGP card is, in addition to not knowing which Opteron's memory controller is closest, and therefore the proper addresses range to use for best AGP performance.
But in some earlier versions of ATI's fireGL driver set, some extremely optimistic (or forward looking) ATI driver hacker decided to include support for a non-pci-device-0 AGP slot. This particular driver set works just fine with dual Opteron memory controllers and AGP.
In the end, the "good" boards are go. In fact, my friends have informed me that the "good" boards will arrive on the 22 of September. If Nvidia does not have their drivers prepared by then, it could be an AMD-VIA-ATI workstation sweep, instead of the beautifully (ruthlessly) competitive market that we've all come to know and love.
This may all seem a bit esoteric, but hopefully it will put to rest the hot debate between the hundreds of bright boffins on Ace's general forum, the source that prompted my inquiry into the subject. You gents will not have to wait much longer for the "perfect" Opteron workstation. ยต