nForce4, PCI Express, and DDR2 for AMD CPU?

FlipCody

Limp Gawd
Joined
Feb 28, 2003
Messages
249
Sorry if these questions have been raised. What is up with search not working? Are they ever going to fix this?

Any news on nForce4? It seems weird to me that Intel, Via, and SiS have chipsets supporting PCI Express coming out in the next few months and one of the major people behind PCI Express (nVidia) does not have a chipset anounced that supports it.

Does anyone know if upcoming AMD chipsets will be supporting DDR2? It does not look like VIA will have it with their upcoming K8T890 chipset. This is weird because it appears that VIA and SIS will have DDR2 support with thier upcoming Intel based chipsets. I wonder if there will be DDR2 support in nForce4.

Thanks,
Flip
 
NF3's are having major problems right now. Anything relating to DDR2 is covered under NDA (all info I have anyway) so can't tell you anything about that.

i think the newest AMD64 chipsets are still too immature to worry about the N4's (ie: they have to correct inherent problems in N3's first before they can worry about n4's or risk losing a lot of R&D time and money)
 
Heh, you DFI guys kick ass (looks down at free shirt I got). But, from what I remember from a talk y'all did at PDXLAN, nForce3 is going to support PCI-Express, but not necessarily with the next wave out (eg. Pro/NP 250).
 
Originally posted by Angry_Games
NF3's are having major problems right now. Anything relating to DDR2 is covered under NDA (all info I have anyway) so can't tell you anything about that.

i think the newest AMD64 chipsets are still too immature to worry about the N4's (ie: they have to correct inherent problems in N3's first before they can worry about n4's or risk losing a lot of R&D time and money)

:rolleyes: major problems right now?

Might want to read up before you post shit.

Nforce3's have locked PCI/AGP and overclock better than the KT800 as well as being just as stable as them.
 
Originally posted by neoreturns
:rolleyes: major problems right now?

Might want to read up before you post shit.

Nforce3's have locked PCI/AGP and overclock better than the KT800 as well as being just as stable as them.

actually there is much controversy if the NF3 actually does.. If you read up on either lostcircuits.com and anandtech.com they tested the PCI bus speeds on NF3's (and a few other boards) and the nf3 does NOT appear to be locking the pci bus... actually, none of the amd64 boards appear to be doing this at this time.. which would explain the shitty OC's that ppl have been routinely getting with the a64's...

And before you start saying ppl are full of shit, you should check their credentials... Angry_Games not only works for DFI, he's one of their main tech support ppl... And if there are inherent problems with the NF3, he'd probably know.. ;)
 
DDR2 would have to be supported by the A64 memory controller, not the chipset. The controller is on the processor die, ensuring the lowest latencies possible. The first DDR2 we've seen (on video cards - 5700Ultra, 9800Pro 256MB) had increased latencies to attain their higher absolute clock speeds, decreasing the performance differential.

PCI-E will undoubtedly be supported by future A64 chipsets, but the first video cards on PCI-E will probably not take much advantage of the interface if they are otherwise identical to the AGP part.
Nvidia will be initially using an add-in chip to convert AGP cards to PCI-E compatible, meaning there should be little or no performance gain possible.
 
both will probably launch bridged solutions as a stopgap measure until they can bring native chipsets to market. plus the bridge chip could improve performance by lowering the latencies. it would essentially just be doing the job of the gart and if it is done properly it could help i/o performance, but as it stands none of the current cards even can saturate agp 8x, so pci-e 16x bandwidth will be a nonissue.
on topic, I expect that amd will wait for ddr2 to become widespread and affordable before moving to it. that way, intel can fight the battles to make ddr2 accepted, and amd can reap the benefits.
 
plus the bridge chip could improve performance by lowering the latencies. it would essentially just be doing the job of the gart and if it is done properly it could help i/o performance, but as it stands none of the current cards even can saturate agp 8x, so pci-e 16x bandwidth will be a nonissue.
How could an additional piece of logic converting a parallel protocol to serial decrease latencies per se without some sort of buffering scheme?

PCI-Express is functionally different from AGP since it is serial, so while the AGP 8X bandwidth may be substantial, IIRC it is limited to either in or out at any given instant, like PATA.

A converter might allow simultaneous transactions to be done outside of the board, but if the board is internally parallel the benefit of doing so would seem to be almost entirely wasted.
 
well of course you have to buffer it, and run the bridge at speeds high enough to do the as fast or faster then the nb gart handles them, that way you won't have any noticible overhead, and by putting the gart functionality on the card itself you can lower the latencies between the gart and the gpu as they will no longer be limited to 66 mhz. it really is a nonissue though as the only real benifit all this would give is more bandwidth and the cards aren't exactly bandwidth starved atm.
 
Back
Top