confused about the different pci-e standards...

Mizugori

[H]ard|Gawd
Joined
Mar 25, 2004
Messages
1,240
Okay I have been out of the loop for a while, now coming back and I'm a little confused about pci-e. I was looking at some Asus p8z68-v mobos and I saw this one that says it has 2 pci-e 2.0 slots (1 4x and 1 16x): http://www.newegg.com/Product/Product.aspx?Item=N82E16813131773

But then some of the more expensive ones say they have 1 4x slot...

And then a few say "2 x PCI Express 2.0 x16 slots (single at x16 or dual at x8 / x8 mode) 1 x PCI Express 2.0 x16 slot [black] (max. at x4 mode, compatible with PCIe x1 and x4 devices)"

What the heck does that mean? Can anyone explain this to me? Sorry, can't believe how quickly you get out of touch if you stop following this stuff even just for several months!
 
Some boards have switches to split the 16 lanes that the CPU provides into x8x8 for GPUs. Other boards, as a cost saving measure, don't have the switches and so can only provide the 16 CPU lanes to a single slot, with the other slot using the 4 lanes from the PCH, so they offer x16x4. Of you intend to do CrossFire/SLI you should try to get a x8x8 board (although it isn't required) - if you have no intention of running multiple GPUs, then the x16x4 boards would be fine.

That is different from the physical configuration of the board however, most boards will still have 2 x16 slots, along with a couple of x1 slots - the difference is in how they are electrically connected.

There are also Gen 3 boards out now that will support PCIe 3.0 when a compatible CPU (Ivy Bridge) is installed. That is less important, since PCIe 2.0 still provides plenty of bandwidth for current GPUs. Doesn't hurt though, if the price difference isn't a problem.
 
The x number refers to the number of lanes available to a PCIe card and is directly tied into the number of connectors on the bottom of the card. Each lane provides bandwidth, so an 8x slot can provide twice the bandwidth of a 4x slot. Graphics cards tend to use a lot of lanes (they use lots of bandwidth) while something like a network card only needs 1 lane so it has a small connector.

The connectors are laid out so that the connectors for a 1x slot match the first connectors of the 4x slot, which in turn match the first set of connectors on an 8x slot and so on - this means that a 1x or 4x card can be plugged into the longer 8x or 16x slot just fine, though it will only be able to run at 1x or 4x speed. The reverse is sometimes true - a smaller slot, say 4x, can be used for an 8x card as long as it's an 'open' slot - this means the end of the slot isn't capped with plastic and the rest of the board is clear of obstructions, allowing the extra connectors on an 8x card to just overhang. The 8x card *should* run at 4x speed, though for some hardware this is inpractical so it may just refuse to work.

Where it may get confusing is that the total number of physical connectors on the motherboard doesn't have to match the number of lanes the chipset/processor supports. So lets say your chipset/processor has enough bandwidth for 20 pcie lanes. The motherboard can be designed so that it can switch between 16 lanes for the first slot and 4 for the second, to sharing the lanes equally between the two (8x each). So both connectors may be the large 16x variety (or a 16x and 8x), but it's electrically impossible for both to be run at maximum bandwidth at the same time.

This sharing system is useful for graphics cards - you generally want as much bandwidth as possible for your GPU, so if you have 1 video card you want to be able to use the full 16 lanes for it. But if you want to run a crossfire/SLI setup with 2 cards you'd want a better balance - so it just switches to 8 active lanes for each card. Right now though cards should be happy on 8x, so it's not a big deal. Just make sure your board can support the second 8x slot (some only have one slot with more then 4x lanes, making it unsuitable for 2 cards).
 
Back
Top