MarkTime wrote:
I asked about this early on, and they said it will support AGP.
I hope so, but it's going to be hard. The Marvell 6436x series do not have any support for AGP.
64bit 66mhz PCI is AGP equivalent, in terms of throughput but you find AGP cards on the market, and not many 64bit 66mhz PCI graphics cards,
If such cards exist the demand for them must be vanishingly small. How many systems implement a 64-bit 66MHz PCI slot and don't have AGP?
No doubt they'd cost for more than the Peg2 itself, and whatever chipset they use would be well out of date.
Apple has used 32-bit 66MHz cards in the past, but I don't know if these are available on the open market.
so they would need to do some hardwdare implementation to give you a physcial AGP slot, as well as whatever else they need to interface that to the marvell.
And that's where the difficulty lies. The only real way to do this is to build an AGP-PCIX bridge chip, but that way is fraught with serious problems. AGP supports things like sidebanding and isosochrinous transfers, that the PCI bus cannot support. There's also issues like voltage conversion, becuase PCI is 3.3v and the latest AGP cards are 1.5v, and saturation of the PCI bus by AGP traffic.
The old 'Socket 7' PC chipsets used this approach to kludge AGP into an architecture that didn't support it, and anyone who's had to deal with such systems knows how poorly that worked.
my assumption is, whatever AGP slot they develop, it will be a 2xAGP slot equivalent (that wouldn't stop someone from supporting higher multipliers, just wouldn't get any more throughput than the marvell can handle)
I think the only hope of getting real, reliable AGP onto the Peg2 is if someone has developed a PCIX-Hypertransport bridge. Putting Hypertransport in there would allow to designers to use one of the AMD, VIA or nVidia AGP controllers that have been designed to work with Opteron and Athlon64 systems.
## Doc