MoBo with maximum amt of PCIe lanes

I've started to prefer the motherboards that have the highest number of PCIe x16 slots. In general a PCI x16 slot is more usable than is a PCIe x1 slot as it can fit most PCIe cards.

The problem is that even though the card features say 7 full-length PCIe slots they all don't necessarily have full-x16 lane bandwidth. As I understand it correctly this may not be a big problem as most expansion cards will accept fewer lanes than they are designed for (even though the may not operate at full capacity).

But as a user I don't care about that, I want as many lanes as possible as long as cost don't become a problem.

I have read from the specs that the Intel X58 gives 40 PCIe 2.0 lanes and the AMD 890FX gives 42 PCIe 2.0 lanes, which to my knowledge are the most powerful chipsets on the desktop/gamer market. It may be ok if you only use one GPU card but may be somewhat problematic if you want to use CrossFire or SLI. If you want x4 SLI there won't be enough lanes to run it at full speed (as 4x16 lanes would be needed for that).

So what solutions are available to provide more lanes for such configurations? Are there more powerful chipsets? Is it possible to use two northbridges to yield more lanes? On what chipsets is that possible? What measures are taken to increase the number of lanes on a motherboard?

I saw that EVGA has a motherboard that has 7 x16 slots and supports 4 way SLI. They use a chipset called Intel Q965. I don't know how that works.

It would also be nice to know how a SLI/CF setup is affected performance-wise if the PCIe x16 slots are reduced to 8 lanes each. Maybe the performance difference will be small.
7 answers Last reply
More about mobo maximum pcie lanes
  1. The total lanes is of course an issue, but more importantly it's how they divided and the overall architecture.

    You don't want to run any current high-end GPU on an x4 lane, but an x4 is perfectly fine for say a RevoDrive, and an x1 for a sound card.

    The AMD 890FX does not outperform the X58, and typically once you go 4-WAY you're looking at x8/x8/x8/x8 and even the 2-WAY GTX 5XX running on x8/x8 will not saturate until you hit larger format 2560 X 1600 or multiple HD 1920 X 1080 monitors. Next is the architecture, the fact is the Intel is the better choice in gaming, and in the vast majority of 'computing' situations.

    I have an EVGA running 3-WAY and 'I wish' I was x16/x16/x16 but I'm not it's x16/x8/x8.
  2. I looked into the manual of EVGA's flagship motherboard; the SR-2. It uses dual Xeon and probably two Intel 5520 that has 32 lanes each but I couldn't see it in the documentation and the components are hidden underneath a huge cooling block. But it seems that it has 56 lanes dedicated for the PCIe slots because the page for PCIe configuration looks like this:

    PCIe1 : x16/x8
    PCIe2 : x8
    PCIe3 : x16/x8
    PCIe4 : x8
    PCIe5 : x16/x8
    PCIe6 : x8
    PCIe7 : x16/x8

    I interpret this as that slot 1, 3, 5 and 7 can either run 2-way SLI at x16 or 4-way at x8. I take it that the motherboard will automatically detect the cards inserted and adjust the speed accordingly. So for example if I have a 2-way SLI on slot 1, and 3 and insert for example an x8 I/O card in slot 5 or 7 it looks like that the motherboard will reduce the SLI to x8 so as to reserve bandwidth for the I/O card. He11, I suspect that this will happen even with an x1 card inserted to one of these slots! But if I instead use slot 2, 4 or 6 for additional cards then the SLI won't be affected at all.

    If the motherboard was a little bit smarter it would instead reserve the bandwidth for each slot from a pool of lanes such that each and every lane is capable of utilizing x1-x16 depending on what the expansion card demands and how many lanes that are available. But maybe that would give rise to higher latency.

    However, from this I conclude that there are 56 lanes available for the PCIe slots.
  3. Hmm...I built and SR-2.

    PCIe 1: x16 if PCIe 2 is empty, x8 if not
    PCIe 2: x8
    PCIe 3: x16 if PCIe 4 is empty, x8 if not
    PCIe 4: x8
    PCIe 5: x16 if PCIe 6 is empty, x8 if not
    PCIe 6: x8
    PCIe 7: x16
  4. So that means that it offers 64 lanes. The mystery of it all is that the Intel 5520 chipset which the SR-2 uses only has 36 PCIe lanes. I think I have figured out how the MoBo can yield so many lanes. I read the following article:,1768-9.html

    So I suspect that the SR-2 uses the technology from the Skulltrail project to split 32 of the lanes into 64. It is similar to SAS/SATA port extenders that split physical channels into several logical channels. So even though there are more lanes for the GPUs to play around with, the bandwidth will be the same as in a configuration using only x8 slots and I seriously doubt that this will give any significant increase in performance. This motherboard is likely to choke at the same display resolution / I/O Intensity level as a regular x8 4-way SLI motherboard.

    I saw that the LGA2011 socket is coming this autumn. The LGA2011 based chipsets will offer 40 lanes which is not a big increase but the lanes will use the PCIe 3.0 specification which is supposed to be twice as fast as the PCIe 2.0, according to the pci-sig website:

    I also spotted this motherboard from Supermicro. It even sports an LSI 2008 SAS/SATA controller giving 8 additional SAS/SATA 6 Gb/s ports. You can see on the picture that it hosts two SFF-8087 ports. The SR-2 has a tacky el cheapo JMicron controller to give additional SATA ports. This makes the Supermicro MoBo look like it gives a lot more value for the money than are the SR-2.
  5. Someone did this; I wouldn't have wasted my time, but so be it. The SR-2 uses the nVidia NF200 co-chipset. The X68/LGA 2011 really should have implemented the PCIe 3.0 which would have effectively doubled the bandwidth in other words e.g. 40 lanes PCIe 3.0 would yield the same as 80 lanes of PCIe 2.x. I have heard all sorts of PCIe lanes #'s for the X68/LGA 2011 and I'm certain the NF200 or its equivalent will be added.

    Odd slot use only Even slots added
    --------------------------- -------------------
    X16 -- NF200 -- X16 (Through NF200) -- X16(X8) SLOT 1
    | |
    | ------------- X8 SLOT 2
    ------- X16 (Through NF200) -- X16(X8) SLOT 3
    ------------- X8 SLOT 4

    X16 -- NF200 -- X16 (Through NF200) -- X16(X8) SLOT 5
    | |
    | ------------- X8 SLOT 6
    ------- X16 (Through NF200) -------- X16 SLOT 7

    Supermicro produces good boards, but I'm not aware of anyone 'smart' putting in on water or doing what has been done to the SR-2. In either case, the SR-2 is 'not' an ideal gaming rig, nor is any dual CPU MOBO. The one I built was over $12K. If you put it on water you can squeeze in: RAID, PhysX, Sound, etc. Again, it's not a 'good' gaming solution.
  6. I appreciate your post. I didn't know that there was such a thing as the NF200. Things like this are not explicitly documented so it is difficult to find information about it unless you ask someone who already knows. It's really difficult to formulate appropriate search terms in Google that lead to the relevant information. In fact Google has a literally chaotic behavior when it comes to search results, i.e. small changes in search terms give rise to a big difference in search results.

    The NF200 still means that 64 "logical" lanes share only 32 physical lanes using some kind of a multiplexer circuitry or "port extender". So it's difficult for me to see that this gives any significant performance increase over a conventional 8x PCIe 4-way SLI setup.

    The PCIe 3.0 sure looks impressive as it appears to be able to offer twice the bandwidth but I'm still sceptical about it. Even though PCIe 3 is going to replace PCIe 2 there will be a considerable amount of PCIe 2 hardware circulating for quite some time to come. So the question is what happens to a PCIe 3 slot when say a PCIe 2 x8 card is inserted: Will the motherboard allocate 4 PCIe 3 lanes multiplexed into 8 PCIe 2 lanes or will it allocate 8 lanes at the reduced PCIe2 bandwidth? If the latter will happen then it is not so impressive and may lead to problems due to lane limitations.
  7. The PCIe 3.0 is 'supposed' to be backwards compatible much like PCIe 2.x is today. I seriously doubt it will offer 'varying' bandwidth like you suggested. My assumption is it will act much like it does today but with double the bandwidth.

    PCIe 3.0

    Things change and until I see a MOBO implementing the PCIe 3.0 I all I can do is guess.

    BTW - the ways that 'I" verify are by direct contact with the MOBO Mfg and then I use GPUz to verify.
Ask a new question

Read More