I'm glad you referenced that article from Sun, SP. I learned a lot from it myself, but I obviously didn't arrive on this scene early enough to reference it for everyone else's benefit.
Most of you seem to be using 760MPX boards, which complicate the performance situation on account only the 64-bit/66 MHz PCI bus is a primary PCI bus, while the 32-bit/ 33 MHz slots are a secondary bus provided by the PCI-PCI bridge included in the 768 southbridge.
http://www.amd.com/us-en/Processors/TechnicalResources/0,,30_182_873_4366,00.html
I can't *actually* say what effect this arrangement may have on cards installed in 32-bit slots on MPX boards, but a little speculation never hurt anyone... so I'd say that normal cards perform better in the 64-bit slots simply because the 32-bit slots are connected to the northbridge second-hand
through the 64-bit/66 MHz bus and the southbridge.
I have a pci geforce 2 mx200 and I use it in the 64 bit slot... it will play max payne perfectly... I haven't done a benchmark on it... but when I put it in a 32 bit slot.. it will barely.. barely run diablo 2.
I'm guessing it uses the extra bandwith.
Your GeForce isn't performing 64-bit data transfers, since that is what the extra length of the connector is for. It might, however, be capable of 66 MHz operation. Indeed, that wouldn't be too surprising-- since it was born and bred in the AGP era. The manufacturer should be able to tell you whether the card supports 66 MHz or not.
I am tempted, though, to use your situation as evidence for my earlier thought that the performance delta is usually due to the 32-bit slots on MPX boards being sub-optimal.
This does not appear to apply to 760MP boards, with their 33 MHz operation accross the board (so to speak). The overview of the 760MP at AMD.com indicates that only one PCI bus is used-- and seing as how I own one, I can attest to the lack of a PCI-PCI bridge device listed anywhere such things ought to be listed in Windows. Being 33 MHz and keyed for 5v signaling, most likely any normal PCI card you care to try will work in any slot on a 760 MP. It's just that some of the slots happen to have the extra connector for 64-bit transfers. No biggie. =)
This is no consolation to those of you with MPX boards whose 64-bit slots refuse to accept certain cards keyed for 3.3v, and therefore should-- according to PCI specs-- function properly. I don't know what to tell you.
As we all know by now, 66 MHz slots are required by the PCI specs to use 3.3v signaling. Oddly, Tyan's relatively new S2468 Thunder K7X features one 66 MHz slot keyed for 3.3v, while the other is keyed for 5v. The board's features proudly boast "One 64-bit 66MHz (5-volt) PCI slot." I don't understand this either, but I see they backed off from this violation on the even newer S2469 Thunder K7X Pro (look at the picture, not the text, on the 2469 page).
www.tyan.com/products/html/athlon.html
On the subject of 64-bit and/or 66MHz cards, I seriously considered a gigabit NIC. Not because I have anything else gigabit to plug into, but because nearly every NIC on the market that features 64-bit PCI also happens to be gigabit. Several cards, such as the Linksys EG1064 and most Intel offerings, are also 66 MHz capable (some will go as high as 133 MHz). Would I see any practical benefit from such a monster? Not in my system, no. But think of how little bus time it would take up! You want to have exclusive bragging rights, right? Incidentally, I went with a 7$ Linksys 10/100 card...
If you really tax your PCI bus, though... say, with a gigantic 64-bit/66 MHz RAID 5 controller, and you also do a lot of network transfers... then I heartily recommend such a beefy NIC to avoid PCI bus saturation and keep all of that *other* expensive equipment humming, regardless of whether there's any gigabit hardware on the network.
Well, based on the length of the post, I hope I managed to be thorough.
