PCI-E, here to stay?

[CoW]8(0)]

I was curious if there was anything on the horizon to replace PCI-E? How long do you think PCI-E will be around for as the mainstream for graphics cards?

 

[Arcygenical]

PCIE 2.0.

Upgrades are imminant .

PCIe 1.0 cards will work in a 2.0 slot though.

 

[WintersFury]

I may be a conspiracy theorist or something, but I really think the only reason that graphics cards really shifted to pci-e in the first place over agp was so they could do dual card solutions and hopefully sell more cards to people.

I've read that a x16 pci-e slot has double the bandwith of an agp 8x. So 8x pci-e which will run cards with no issues and did the dual card thing for some time has enough bandwith so an agp slot had enough bandwith.

So the only reason I have left as bandwith wasn't the issue is that they wanted to do dual card stuff to sell more cards. I don't see pci-e really being a huge thing in the desktop segment as people don't need the bandwith of pci-e 1.0. Though I imagine it'll be foisted upon us if we like it or not by chipset manufacturers much like pci-e 1.0 was.


EDIT:: To better answer the question. Since it's a very scalable interface in terms of performance and adding more slots and such I can't see it going away. If AGP could have been as easially used for other things or had another slot added for dual card solutions and all we may still be using it today. As said it will be upgraded, but I doubt it'll change enough to make the old stuff not work with the new. It's possible as they may be changing clock speeds or something, but I think they'll make sure the new slots work with the old cards due to the fact that there is a lot of pci-e 1.0 stuff out there being used and consumers don't like big changes that make their old hardware not work anymore.

 

[lozaning]

I may be a conspiracy theorist or something, but I really think the only reason that graphics cards really shifted to pci-e in the first place over agp was so they could do dual card solutions and hopefully sell more cards to people.

I've read that a x16 pci-e slot has double the bandwith of an agp 8x. So 8x pci-e which will run cards with no issues and did the dual card thing for some time has enough bandwith so an agp slot had enough bandwith.

So the only reason I have left as bandwith wasn't the issue is that they wanted to do dual card stuff to sell more cards. I don't see pci-e really being a huge thing in the desktop segment as people don't need the bandwith of pci-e 1.0. Though I imagine it'll be foisted upon us if we like it or not by chipset manufacturers much like pci-e 1.0 was.


EDIT:: To better answer the question. Since it's a very scalable interface in terms of performance and adding more slots and such I can't see it going away. If AGP could have been as easially used for other things or had another slot added for dual card solutions and all we may still be using it today. As said it will be upgraded, but I doubt it'll change enough to make the old stuff not work with the new. It's possible as they may be changing clock speeds or something, but I think they'll make sure the new slots work with the old cards due to the fact that there is a lot of pci-e 1.0 stuff out there being used and consumers don't like big changes that make their old hardware not work anymore.

your argument discounts the fact that agp supplied iirc 45 watts to the card, and pci-e around 75. there are other improvements besides bandwidth.

 

[fattmarrell]

I was think kind of the same question today, but instead of which new slots will be available I was wondering how they'll effectively produce more horsepower out of the pcb's.

Pcb size is at its limits, and the heatsinks are at their limits. Temps seem to be hitting limits too. I guess nano size is where it's at? Then multiple gpu's per board?

 

[Sycraft]

The reason for the PCIe move isn't the bandwith, it is that AGP sucks. AGP was a hack from the get go and Intel will freely admit that. Normal PCI was at its limits, 64-bit 66-MHz PCI was WAAAY too expensive on normal boards, something was needed. So along comes AGP. Never intended to last as long as it did.

PCIe is better in four major ways:

1) Support for more than one slot. AGP was inherently limited to a single slot. Now maybe you think who cares, only high end gamers want that. Well no, there are plenty of situations where more cards are useful. It is easy to want to have 3 or more monitors on things like presentation systems. With AGP you either had to get PCI cards, or find a card with more than 2 outs. No such problems with PCIe.

2) Support for bi-directional transfers. AGP is a very one-way design. It's good for getting data to the card, not good for getting it back from it. That was ok when cards first came out, but now we have cards that can do useful calculations and it'd be nice to have fast access to that. PCIe allows that, it is a bi-directionaly fast standard.

3) Because of those two things, AGP was ONLY for video cards, period. Other high bandwidth devices were SOL and had to go with the expensive, and rare, option of 64-bit PCI and PCI-X. PCIe is universal so all cards can use teh same interface and a high bandwidth RAID card works just as well in a slot as a graphics card does.

4) PCIe is serial. PCI was a parallel standard, meaning that all the traces had to be the exact same length. Made motherboards more complicated, especially for the 64-bit variety which has a ton of traces. PCIe is serial, meaning each lane is independent. Makes design and scalability much easier. Thus it is feasible to whack 3 16x slots on a consumer board where as PCI-X you never say it until the board was more expensive.

Really it was a very needed protocol upgrade and too long in coming. Now you can expect it to be with us for quite some time. Thus far it has tons of bandwidth for any use, and as with the 2.0 standard we see it is scaling well. I'd bet it sticks around at least a decade (with upgrades of course).

 

[needsnumbers]

I agree with the above but want to through out a few more reasons for PCIe and why it's here to stay. Biggest reason; hard drive speeds. Until we have non-volitile storage that can keep up with ram, there is no useful reason for improving the bus speed of PCIe from where it is now. The only exception will be when (and if) we get to the point of having such high resolutions that our gpu's consume more $ worth of electricity than the rest of the house combined. Until then, expect PCIe to live a long and happy life. Look at history, ISA was offered on most boards up until around 99 and not until a few years ago did it die out. Considering that run, and it's bandwidth compared to the rest of the system (hard drive specifically), I wouldn't expect PCIe to go anywhere for the next 5-6 years. Even when a replacement comes along, and even if it gets adopted fast, it's more likely you'll see the then "old" PCIe slots move down the boards to push out PCI. Speaking of which, don't think PCI will be gone for another 4 years or more. Companies pushing hardware will get in trouble. We are already in a period of not needing to upgrade our computers because the 2 biggest issues haven't been sorted out yet... hard drive speeds and operating systems. Until those 2 catch up, don't expect much change.