PCI Express 2.0 vs. 1.x compatibility?

[legocitytruck]

I am upgrading a few year old Dell gx620 with a dual output video card to use two monitors. The specs on the machine list it as having a PCI Express x16 slot, but I don't know if it is 1.0 or 1.1. If the slot is 1.0, will any PCI Express 2.0 card work (of course with a slight bottleneck) or does the slot have to be 1.1 to use any PCI Express 2.0 card? Does anyone have any first hand experience with this?

 

[legocitytruck]

I just learned per dEll's website that the slot is PCI Express 1.0a, will any 2.0 card work?

 

[Stumps]

yes, 99.99% of PCI-E 2.0 cards will work perfectly fine, assuming your PSU is up to scratch.

most PCI-E 2.0 cards will not be bottlenecked by the PCI-E 1.x slot.

 

[CZroe]

Originally posted by: Stumps
yes, 99.99% of PCI-E 2.0 cards will work perfectly fine, assuming your PSU is up to scratch.

most PCI-E 2.0 cards will not be bottlenecked by the PCI-E 1.x slot.

Actually, none will. Just like AGP, the interface standard is moving forward well in advance of the actual need. It's why x16 PCIe slots with x8 electrical connections are fine and dual x16 was considered "overkill" when ift first arrived. "Most" today still wouldn't be bottlenecked by AGP 8x, bandwidth-wise.

 

[Nathelion]

You are wrong sir. Modern dual-GPU cards are bottlenecked by a PCI 1.x 16x slot.

 

[Scali]

Aside from hardware bugs, I believe PCI-e is both backwards and forwards compatible.
So a PCI-e 2.x card will work in a PCI-e 1.x slot, and a PCI-e 1.x card will work in a PCI-e 2.x slot (obviously performance will be lowest common denominator).

As for the 1x, 4x and 16x variations... Physically, they are backwards compatible, so a 1x and 4x card will work in a 16x slot. But since a 1x and 4x are physically smaller slots, you can't insert larger cards.
There are motherboards with physical 16x slots, while electrically they may only be 1x or 4x. Generally 16x cards will just work, albeit at reduced performance obviously.

 

[CZroe]

Originally posted by: Nathelion
You are wrong sir. Modern dual-GPU cards are bottlenecked by a PCI 1.x 16x slot.

I'm not talking about bandwidth between GPUs but, rather, bandwidth required between the GPU and the system. Yes, raw transfer speed is bottlenecked in the same way that ATA100 interfaces "bottlenecked" ATA133 drives... for a fraction of a second until that 2-8MB buffer was exhausted and you were bottlenecked by the physical medium limitations. We aren't talking about raw transfer speeds either.

For example, your overall gaming/3D performance would be bottlenecked by the CPU and FSB/memory performance of an AGP system before the AGP8x slot's bandwidth had a chance to be a limiting factor. One could CLAIM that an AGP8x card was "bottlenecked" in an AGP4x slot, but we all know that just because it could use the maximum bandwidth for moments doesn't mean it truly benefited from AGP8x, needed it, or was bottlenecked without it. We already know that the bus is just used to get the data to the card and the on-board memory and GPU take it from there. If there isn't enough memory on the card and the full bandwidth of the bus is needed for every rendered frame, performance would be abysmal no matter what bus/interface you were using. The point is, don't hit the bus except for initial loads and have enough bandwidth in that bus to feed/supply the card new textures and geometry in real-time. 16x PCIe 1.0 has that in spades. The nature of SLI/Crossfire puts more importance in the bus than simply loading what is needed onto the card and performing operations from there, so multi-GPU is out of context. For example, you couldn't throw an AGP bridge chip on a multi-GPU card to compare because the PCIe bus is being used for communication between GPUs and not just the system.

 

[Stumps]

Originally posted by: CZroe

Originally posted by: Nathelion
You are wrong sir. Modern dual-GPU cards are bottlenecked by a PCI 1.x 16x slot.

Raw transfer speed is bottlenecked. We aren't talking about that. For example, your overall gaming/3D performance would be bottlenecked by the CPU and FSB/memory performance of an AGP system before the AGP8x slot's bandwidth had a chance to be a limiting factor. One could CLAIM that an AGP8x card was "bottlenecked" in an AGP4x slot, but we all know that just because it could use the maximum bandwidth for moments doesn't mean it benefited from AGP8x, needed it, or was bottlenecked without it. We already know that the bus is just used to get the data to the card and the on-board memory and GPU take it from there. If there isn't enough memory on the card and the full bandwidth of the bus is needed for every rendered frame, performance would be abysmal no matter what bus/interface you were using. The point is, don't hit the bus except for initial loads and have enough bandwidth in that bus to feed/supply the card new textures and geometry in real-time. 16x PCIe 1.0 has that in spades.

have a read

http://alienbabeltech.com/main/?p=2249

 

[CZroe]

Originally posted by: Stumps

Originally posted by: CZroe

Originally posted by: Nathelion
You are wrong sir. Modern dual-GPU cards are bottlenecked by a PCI 1.x 16x slot.

...

have a read

http://alienbabeltech.com/main/?p=2249

I wasn't done writing and F'd it up at that point when it was accidentally posted. Read again.