[SOLVED] Newb Question - Using multiple PCIe 16x slots (and Z690 vs Z790)

[trajan]

I saw a recent question about whether using the second (lower) PCIe slot would reduce graphics performance, but I have a different question. And sorry if this is a stupid one:

Does using the lower PCI x16e slot (in a motherboard that has one) risk slowing down the top (GPU) slot and GPU performance?

I'm splurging on a high end build: ASUS Formula Z690, RTX 4090, i9 13900k. There's a note on the ASUS site regarding the Forumula, though, and I suspect this isn't unique to this board only:

Intel® 13th & 12th Gen Processors*

2 x PCIe 5.0 x16 slot(s) (supports x16 or x8/x8 mode(s))**
Intel® Z690 Chipset***
1 x PCIe 4.0 x16 slot (supports x4 or x4/x4 mode)***
* Please check PCIe bifurcation table in manual Chapter 1
** When ROG Hyper M.2 card is installed on PCIEX16(G5)_1, PCIEX16(G5)_2 will run x8 only and if ROG Hyper M.2 card is installed on PCIEX16(G5)_2, PCIEX16(G5)_1 will run x8 only.
*** Supports Intel® Optane Memory H Series on PCH-attached PCIe slot

I'm not planning to run two GPUs ever, so the RTX 4090 will go in the top slot, but I like the idea of the lower slot as potential future proofing. Specifically, it would be comforting to know if I could use that slot in the future for PCI 5.0 SSDs (that should be possible, right?). But the note above seems to be saying that if I use one PCI x16 slot, it will cause the other slot to run at x8. I know this is specifically relating to this ROG Hyper M.2 card, but would the same apply to any high-bandwidth card that I put in the lower slot? In other words, is there limited bandwidth that's shared between the two PCI x16 slots, and using both will limit the bandwidth available to the GPU in a way that might affect performance?

I'm aware that I'm making a small trade off on future proofing by going with the a Z690 board now that Z790 are available, but from what I can tell as a novice, there shouldn't be much/any performance difference at stock speeds and probably not a big difference on OC performance except at the higher ranges. The Formula matches the white aesthetic I want, but doesn't have PCI 5.0x M.2 slots. That speed seems excessive anyway for my use case (gaming/office), but it would be comforting to know if it is possible to upgrade later using my second x16 slot.

Thank for any advice/reactions, I'm psyched for this build but feeling very ignorant right now.

 

[kanewolf]

I saw a recent question about whether using the second (lower) PCIe slot would reduce graphics performance, but I have a different question. And sorry if this is a stupid one:

Does using the lower PCI x16e slot (in a motherboard that has one) risk slowing down the top (GPU) slot and GPU performance?

I'm splurging on a high end build: ASUS Formula Z690, RTX 4090, i9 13900k. There's a note on the ASUS site regarding the Forumula, though, and I suspect this isn't unique to this board only:



I'm not planning to run two GPUs ever, so the RTX 4090 will go in the top slot, but I like the idea of the lower slot as potential future proofing. Specifically, it would be comforting to know if I could use that slot in the future for PCI 5.0 SSDs (that should be possible, right?). But the note above seems to be saying that if I use one PCI x16 slot, it will cause the other slot to run at x8. I know this is specifically relating to this ROG Hyper M.2 card, but would the same apply to any high-bandwidth card that I put in the lower slot? In other words, is there limited bandwidth that's shared between the two PCI x16 slots, and using both will limit the bandwidth available to the GPU in a way that might affect performance?

I'm aware that I'm making a small trade off on future proofing by going with the a Z690 board now that Z790 are available, but from what I can tell as a novice, there shouldn't be much/any performance difference at stock speeds and probably not a big difference on OC performance except at the higher ranges. The Formula matches the white aesthetic I want, but doesn't have PCI 5.0x M.2 slots. That speed seems excessive anyway for my use case (gaming/office), but it would be comforting to know if it is possible to upgrade later using my second x16 slot.

Thank for any advice/reactions, I'm psyched for this build but feeling very ignorant right now.

You would NOT get PCIe 5.0 performance. The data you posted show that the third slot is PCIe 4.0 x4.

 

[DRagor]

The top two x16 slots are wired directly to CPU, and share its 16 5.0 lanes. Whatever you put into second slot, no matter what it is, will cut the top slot to 8 lanes. Basically it is good option for future mid range GPUs that will be wired 8 x5.0 (same way like Radeon 6600XT is wired 8 x 4.0) as it will let you use top slot for full potential with such GPU and use bottom slot for two M.2 5.0 drives.

 

[trajan]

The top two x16 slots are wired directly to CPU, and share its 16 5.0 lanes. Whatever you put into second slot, no matter what it is, will cut the top slot to 8 lanes. Basically it is good option for future mid range GPUs that will be wired 8 x5.0 (same way like Radeon 6600XT is wired 8 x 4.0) as it will let you use top slot for full potential with such GPU and use bottom slot for two M.2 5.0 drives.

I think I follow... So bandwidth is shared between the two x16 slots on the motherboard. If only PCI 5.0 x8 becomes available to the GPU (top slot) due to use of the bottom slot, how would I figure out if that's going to be limiting to GPU performance or not? I had assumed that the RTX 4090 needs x16 (full), but poking around at the specs it seems like its designed to run on PCI 4.0. So maybe having PCI 5.0 x8 (instead of x16) won't actually limit its bandwidth?

 

[trajan]

You would NOT get PCIe 5.0 performance. The data you posted show that the third slot is PCIe 4.0 x4.

Sorry for being imprecise here! I'm asking specifically about the two PCI 5.0 x16 slots on the motherboard, not the other PCI 4.0 x4 slot.

 

[DRagor]

I had assumed that the RTX 4090 needs x16 (full), but poking around at the specs it seems like its designed to run on PCI 4.0. So maybe having PCI 5.0 x8 (instead of x16) won't actually limit its bandwidth?

It will. The slot is 5.0, but GPU is 4.0, and it is always lower speed of the two parts being used. Of course it is not a killer thing, probably a few % (haven't checked if anyone on youtube run the tests yet) but still.

 

[trajan]

It will. The slot is 5.0, but GPU is 4.0, and it is always lower speed of the two parts being used. Of course it is not a killer thing, probably a few % (haven't checked if anyone on youtube run the tests yet) but still.

Thanks for taking the time to explain this, I really appreciate it and I think I understand it now, or at least well enough - I think! Using both slots will split the bandwidth between the two slots - both will run at PCI 5.0 speeds, but with half the lanes (bandwidth), so x8 each. Whether that limits GPU performance or not will depend on how much bandwidth the specific GPU needs.

I did some searches based on this understanding and found this article, which seems on-point and verifies your hunch perfectly in terms of performance for the RTX 4090. https://www.techpowerup.com/review/nvidia-geforce-rtx-4090-pci-express-scaling/29.html

They were trying to see if running at PCI 3.0 speeds or running at PCI 4.0 x8 (instead of x16) made a difference, and they found only a negligible difference, a few % just like you guessed. I would think that there's no reason for the difference between PCI 5.0 x8 vs x16 to be any greater (in should be less, if anything).

Thank you again for the advice!