PCI Express 3.0 vs. PCI Express 2.0

First of all this is my first post and am happy to be here and excited to be a part of the DS community.

I have a question that is more to do with learning about my graphics card and the PCI 3.0 tech and less about finding a work around which I can figure out on my own.

Let me say that I am coming from an old AGP rig, so though I follow tech news closely, I have zero experience with the newer tech. (FYI for the kids out there, AGP is not a movie about aliens and/or predators.)

So let's get to it.  I have an ASUS Z97M-PLUS mobo with an AMD Radeon (Sapphire) R7 260X. The mobo includes 1 x PCIe 3.0/2.0 x16 (x16 mode, gray), 1 x PCIe 2.0 x16 (x4 mode, black), 2 x PCI
I have a PCI (not PCIe) wifi card (Rosewill RNX-N150PC) that I want to install in my new machine but there is no room due to the size of the v-card. It completely covers one PCI slot and covers enough of the second one to make it impossible to use.  I fully expected the former slot to be inaccessible but was planning to use the latter when I ordered my rig.

So know I am wondering if I could just move the GPU to the second PCIe slot (that I am going to eventually use for a Crossfire config).
I know I can just get a USB wifi adapter as a workaround, but again I am a curious one.

The mobo manual notes: "In single VGA card mode, use the PCIe 3.0/2.0 x16_1 (gray) for a PCI Express x16 graphics card to get better performance."

So my question is this: What kind of performance difference are they talking about?  Also, when I do decide to beef this thing up and add another GPU, should I not bother buying a PCIe 3.0 card and stick with an older PCIe 2.0 card?

Thank you!

FYI: I am also rocking an i5 4690K with 8GB of RAM (1866MHz).

 be careful populating the second physical x16-sized slot (the x4 slot on your board) and not populating the first.  a lot of boards split the lanes when the second slot is occupied, even if there isn't a card in the first slot.  now, there really is no noticeable difference in performance between x16 and x8, i think the throughput speed is equal to half the slot lane (so, x16 = 8Gb/s, x8 = 4Gb/s, etc.) in PCIe 2.0.  PCIe 3.0 effectively doubles the lane bandwidth relative to 2.0,  so you're back to almost 1GB per lane. 

when you mix two generational standards, as in the case of your mobo, it functions at the lowest communication standard that both can use, PCIe 2.0 in your case.  so if you use a 2.0 card in a 3.0 system it will function as 2.0.  what this means is that you need a cpu/mobo/gpu(s) all on the 3.0 standard to function at 3.0 speeds. 

as far as real performance hit is concerned in sli or crossfire, going from x16/x16 to x16/x8, or x8/x8 isn't going to be noticeable in either PCIe 2.0 or 3.0.  going to x16/x4 or x8/x4 or x4/x4 will be noticeable (although not horrible, unless you're the type that thinks a 5% drop in performance is the end of the world) depending on the game, resolution and settings.  populating the x4 slot (even with a 3.0 card) means you'll be getting 2Gb/s throughput.  you'll probably see more frames dropped trying to dual card that mobo. 

for crossfire's latest generation of cards they can operate together as long as they are of the same series, in your case two R7's.  sli generally needs two cards with the same gpu (i.e. 2xgtx 970) to function properly. 

i/m sure Meller or somebody will correct me if i've gotten any of this wrong. but this is my understanding of what it means for you. 

Originally posted by db188 going from x16/x16 to x16/x8, or x8/x8 isn't going to be noticeable in either PCIe 2.0 or 3.0.  going to x16/x4 or x8/x4 or x4/x4 will be noticeable

One more question... you speak of going from x16/x16, but this is not possible for me.  The questions for me is, "What happens when I go from x16 to x16/x4?"

Couldn't get back last night, but to further help...

A lot of testing of x16 vs x8 vs x4 and 3.0 vs 2.0 isn't actually done in those specific slots, but is done by taping the video card's connector to simulate less bandwidth and slipped back into a x16 wired slot. That's why only a 5% or so only loss in performance is shown. Not sure about that mobo, but in a lot of cases, the 2.0 x4 slot is wired to the chipset and not to the cpu directly.   The added task of going through the chipset can cause a 30% + performance drop. If that's the case, it will drop your primary card down to the secondary cards performance so they can work in together, mirror each other. Using the bottom x4 2.0 slot for other types of cards might also disable any other 2.0 slots. But, as db188 said, won't effect your primary card.   Not sure if it disables the older Pci, non e.

This is all mobo specific and I can't say for sure if yours is the case. Info is sketchy.



Edited by bprat22 - 01 Nov 2014 at 3:42am

So, starting at the processor, it has 16 lanes of PCI-e 3.0 coming from it in addition to 4 lanes DMI 2 (which can be reconfigured as 4 lanes of PCI-e 2.0, but that has to be done on the motherboard) connecting to the Platform Chip Hub (PCH). You don't get any more than this on consumer processors; you have to go to enthusiast or enterprise processors to get more.



ASUS have routed all of the PCI-e 3.0 lanes to the 1st slot for the Z97M-Plus. This is the easiest, cheapest, and highest performing configuration. When you see boards where multiple configurations are available, there are switches in the middle there, and compensation for impedance and balancing the timing of all of the electrical differential pairs that make up those connections. These cause additional line latency, while in the nanosecond range, is still significant when each tick of the clock on your processors is sub-nanosecond.

The DMI is connected of the PCH on the board; this is under the round ASUS "heatsink." The PCH has switch and translation for the network connection (Ethernet), SATA (storage and optical drives), USB, COM, and has 8 additional PCI-e 2.0 lanes. All of these devices share the bandwidth of the DMI connection to the processor.

On the Z97M-Plus, 4 of the PCI-e 2.0 lanes off of the PCH are sent to the 4th PCI-e slot, while 1 is sent to a PCI bridge for the two PCI slots and 2 to the M.2 slot. The data going through the PCH not only has greater latency, as it isn't direct to the processor, it also has to contend for bandwidth with all of the other devices connected through it. I hope this clears up the underlying structure of the motherboard.

And Meller, yes those charts hold up for the current state of GPUs. But once NVIDIA releases Pascal-based chips, the difference between the bandwidth to and from the cards will play a much more significant role. NVIDIA has hinted at the with their NVLink technology (although I'm stil wondering what the adoption strategy to get CPU manufacturers interested in NVLink will look like). That gives the validity of those graphs a year and a half to two.

I wasn't saying what you posted was incorrect, just that it is only valid for the current and near-future hardware. It will change only the current bottleneck, memory access latency and bandwidth, improves, which you were correct in pointing out stack memory will help with. However, stacked memory was initially in the timeline for Volta, but Volta has been pushed back, but they did move that part of the technology forward to Pascal, which succeeds Maxwell and arrives in 2016. While yes this is "speculation," my intent was to say that bottlenecks in the future will change and the fact of little GPU performance disparity between PCI-e interface bandwidths will no longer be true.

Thanks a lot guys. As it turns out I WAS able to wedge the PCI card in there.  I will have to be extra diligent in cleaning though.  The filters on the case will pay off in spades!

But again this question was more about me understanding the tech and I really appreciate the help. I can now look forward to using that 2.0 slot for a capture card or something.  Thank you for having me here at the DS forums.

Well, it's can't be a club... because as the great Groucho Marx once said: "I refuse to join any club that would have me as a member."