• Welcome to TechPowerUp Forums, Guest! Please check out our forum guidelines for info related to our community.

NVIDIA GeForce RTX 4090 PCI-Express Scaling

AM4 platform doesn't support NVMe 2.0 standard, only up to 1.4, I believe.


It is that simple.
Bifurcation of x16 to x8x8 leaves eight lanes of whatever generation for any device to try to use it.

As no commercial device supports Gen5 speeds, slotting Gen4 device into Gen5 M.2 or PCIe slot will work as Gen4 device.

Eight Gen5 lanes left for GPU in x16 slot after bifurcation will work as Gen4 x8.
PCI-E 2.0, not NVME 2.0
PCI-E 5.0 slots can run at 1,1/2/3/4/5 at x1/x2/x4/8/x16

I wanted to check that a PCI-E 2.0 NVME drive in that 5.0 slot wouldnt lock a GPU to 2.0 x8
 
PCI-E 2.0, not NVME 2.0
PCI-E 5.0 slots can run at 1,1/2/3/4/5 at x1/x2/x4/8/x16

I wanted to check that a PCI-E 2.0 NVME drive in that 5.0 slot wouldnt lock a GPU to 2.0 x8

There is no locking. Bifurcation works via PCIe switch chip, which allows all devices on split slots to work independently up to PCIe standard supported by the switch chip in the first place, then slot, then device.
 
It would be a real shocker if AMD came out with some Gen 5 GPUs that made good use of Pci-Ex Gen 5 bandwidth, and smeared egg on Nvidia's face by taking the performance crown, but I just don't see that happening. Still Blue and Green, just like my Seahawks. ;):D
 
There is no locking. Bifurcation works via PCIe switch chip, which allows all devices on split slots to work independently up to PCIe standard supported by the switch chip in the first place, then slot, then device.
It's done at boot, and cant change til a reboot. It's "locked in"
 
It would be a real shocker if AMD came out with some Gen 5 GPUs that made good use of Pci-Ex Gen 5 bandwidth, and smeared egg on Nvidia's face by taking the performance crown, but I just don't see that happening. Still Blue and Green, just like my Seahawks. ;):D
Because the 4090 is chocked by even pcie 3.0 with 1 outlier lol? Tell us more!
 
Because the 4090 is chocked by even pcie 3.0 with 1 outlier lol? Tell us more!

Trust me, I'd be completely happy to see AMD thoroughly trounce Nvidia on performance. If nothing else it might bring prices down a smidge. The fact is though, at best usually AMD can only come close to matching them, and they're no saints when it comes to GPU pricing either lately on their high end models at launch. Of course they always come down to more reasonable levels when people realize they don't perform as well though. So yeah, LOL all you want bud, in denial much? :rolleyes:

And only a fool would buy a 4090 and put it on a Pci-Ex 3 board, talk about LOLs. I wouldn't even trust Nvidia's cheesy 16 pin adapter myself.
 
Last edited:
Trust me, I'd be completely happy to see AMD thoroughly trounce Nvidia on performance. If nothing else it might bring prices down a smidge. The fact is though, at best usually AMD can only come close to matching them, and they're no saints when it comes to GPU pricing either lately on their high end models at launch. Of course they always come down to more reasonable levels when people realize they don't perform as well though. So yeah, LOL all you want bud, in denial much? :rolleyes:

And only a fool would buy a 4090 and put it on a Pci-Ex 3 board, talk about LOLs. I wouldn't even trust Nvidia's cheesy 16 pin adapter myself.
But as techpowerup proved that there is no statistical significant delta when using pcie 3.0 like if using a z790 board and a next gen pcie 5.0 nvme as well as a 4090 except for the one outlier. Questions?
 
But as techpowerup proved that there is no statistical significant delta when using pcie 3.0 like if using a z790 board and a next gen pcie 5.0 nvme as well as a 4090 except for the one outlier. Questions?

You're only pointing out the obvious there. Every time a new Pci-Ex standard comes out, the new GPUs releasing can't make anywhere near use of it's bandwidth. So what if some of us keep checking if that remains to be true. You want a cookie for being overly pedantic or something? JEEEZ!
 
You're only pointing out the obvious there. Every time a new Pci-Ex standard comes out, the new GPUs releasing can't make anywhere near use of it's bandwidth. So what if some of us keep checking if that remains to be true. You want a cookie for being overly pedantic or something? JEEEZ!
Well the blindside is that AM5 doesn't have that problem so at least its something plus they have zen4 3d chips coming out. So while yes I love for AMD to stick it to Nvidia as much as the next person in reality they are coming close. If you can extrapolate the Zen multi chiplet design from launch to actually becoming a competitive contender it took about 2 generations for everyone to want it besides a few loyalists. I predict we will have something similar with next generation MCM AMD gpus when I say Next generation I mean 7000 rdna3 successor.
 
And only a fool would buy a 4090 and put it on a Pci-Ex 3 board, talk about LOLs.
This is not correct. The loss in gaming of 4090 is 2% on average in x16 PCIe 3.0 slot. it's negligible.
As for productivity apps, we have no measurements at the moment. It's still ok to put Gen4 x16 device into Gen3 xq16 slot.
This might change in 2024, but it's still ok.
 
@W1zzard I was meaning the RPL CPUs and their PCIe lanes.
 
This is not correct. The loss in gaming of 4090 is 2% on average in x16 PCIe 3.0 slot. it's negligible.
As for productivity apps, we have no measurements at the moment. It's still ok to put Gen4 x16 device into Gen3 xq16 slot.
This might change in 2024, but it's still ok.

Yeah it's "OK" so to speak, but anyone going for a 4090 is probably going to want the tech in the Gen 4 MBs, and be able to afford it. That's all I'm saying, no need to get marmy. :rolleyes:
 
Yeah it's "OK" so to speak, but anyone going for a 4090 is probably going to want the tech in the Gen 4 MBs, and be able to afford it. That's all I'm saying, no need to get marmy. :rolleyes:
Like calling people stupid oh wait!
 
@W1zzard I was meaning the RPL CPUs and their PCIe lanes.
Here is entire diagram for Z790 and B760.
Z790.JPG
 
Last edited:
You're good at making diagrams. You should make a thread and post them all in, adding for socket/chipset as you have the time.

Guarantee it'll be popular.
Oh and watermark them, you'll find em appearing all over the web soon enough
 
You're good at making diagrams. You should make a thread and post them all in, adding for socket/chipset as you have the time.

Guarantee it'll be popular.
Oh and watermark them, you'll find em appearing all over the web soon enough
Indeed. I'd love to add such diagrams to our CPU database. @Tek-Check?
 
Thank you. Could you PM me and explain how this would work?
Draw them for all the major CPUs in our database, send to me, I'll add :)
 
I did this in my build because I misunderstood how the M2_1 slot stole lanes. I thought it would drop to PCIe 5.0x8 and not 4.0x8. I have the M2_1 and 2 slots populated with gen 4 ssds in raid 0.

I need to get back into the case because additional storage arrived along with a new psu. This seems like a good time to move the M2_1 slot drive to a different slot. Does anyone know if this would break the array? The raid is Intel rst through the bios.
I just went and did it. Everything works perfectly and I have my 4x16 back. Have not noticed a difference :laugh: Modern motherboards are something else lol.
 
Can anyone help me understand the lane splitting for the Intel vs AMD Asus ProArt Creator boards?

Intel: https://www.asus.com/motherboards-components/motherboards/proart/proart-z790-creator-wifi/

AMD: https://www.asus.com/motherboards-components/motherboards/proart/proart-x670e-creator-wifi/

The AMD says it can run 3 gpus in x8, x8, then x2 for the third slot
Whereas the Intel says it will run x8, x8, x4?

I would NOT be using an NVME in a PCIE5 slot. Either a) an NVME in a gen4 slot, or b) no nvme at all.

If AMD has more lanes, why is the Intel running the third slot at a better speed?
And while a lot of games have been tested, I would use this for 3D rendering, where render power/speed scales linearly with each GPU added. 3 GPUs is 3x faster than 1.

I am just confused about how these two systems are allocating lanes to the slots vs other components.

Thanks!!
 
Can anyone help me understand the lane splitting for the Intel vs AMD Asus ProArt Creator boards?

Intel: https://www.asus.com/motherboards-components/motherboards/proart/proart-z790-creator-wifi/

AMD: https://www.asus.com/motherboards-components/motherboards/proart/proart-x670e-creator-wifi/

The AMD says it can run 3 gpus in x8, x8, then x2 for the third slot
Whereas the Intel says it will run x8, x8, x4?

I would NOT be using an NVME in a PCIE5 slot. Either a) an NVME in a gen4 slot, or b) no nvme at all.

If AMD has more lanes, why is the Intel running the third slot at a better speed?
And while a lot of games have been tested, I would use this for 3D rendering, where render power/speed scales linearly with each GPU added. 3 GPUs is 3x faster than 1.

I am just confused about how these two systems are allocating lanes to the slots vs other components.

Thanks!!
Because the AMD has an entire extra 5.0 M.2 slot?

You wouldnt use GPUs for rendering in any of these 2x or 4x slots they'd be crippled. If you're truly running multiple high end GPU's you're going to want a HEDT board, or at least boards with better slot layouts
1670920777008.png
 
Because the AMD has an entire extra 5.0 M.2 slot?

You wouldnt use GPUs for rendering in any of these 2x or 4x slots they'd be crippled. If you're truly running multiple high end GPU's you're going to want a HEDT board, or at least boards with better slot layouts

Appreciate the insight. I didn't even realise myself until they were next to each other. :)

You can use GPUs for rendering on these boards... why do you say not? PCIE4x4 is the same as 3x8 as seen in the article parent of these comments, and as GPUs (even 4090s) don't saturate PCIE3, it's been functionally shown that high end GPUs don't take a render hit when going from PCIE3x16 down to 3x8 (which is the same as 4x4). It's not frame counts for games I'm talking about it, it's just sending data to VRAM then waiting 10mins for the CUDA cores to chew over it. Once the data is transferred, the PCIE speed doesn't materially affect the calculation of the image - that's down to the CUDA cores and other such things.

Anyway maybe 4x2 would see a hit more so than 4x4, but the question is, is it economically worth it to put the third 4090 on this board and lose 5%, or build an entire second machine with all associated costs (or build a threadripper machine, well that's certainly not economical at all, I'd take the 5% hit rather than spend another $3k on parts).
Even then it's only the 3rd slot taking the performance hit. The other 2 will render 3D as fast as anything else out there.

What I was more interested in was if the AMD board would see a 3rd slot speed of 4x2 with zero NVMEs installed, or how you tell which NVME slot to use that won't affect slot3's speed. But maybe it's better to use the intel board anyway, even if I won't be able to upgrade the CPU in 2yrs.

edit: It seems the AMD will run at x2 no matter what? Even if no NVME installed in that slot. This is what confuses me as I thought AMD had more lanes or at least the same as Intel, the way they "count" and market them is confusing. Someone else said Intel doesn't "count" chipset lanes which is why AMD is listed at 24 but Intel lists 20. But even so why the difference on that 3rd slot... I was simply hoping for "one simple trick" to run that 3rd slot at x4 not x2. :) Ah well.
 
Last edited:
Appreciate the insight. I didn't even realise myself until they were next to each other. :)

You can use GPUs for rendering on these boards... why do you say not? PCIE4x4 is the same as 3x8 as seen in the article parent of these comments, and as GPUs (even 4090s) don't saturate PCIE3, it's been functionally shown that high end GPUs don't take a render hit when going from PCIE3x16 down to 3x8 (which is the same as 4x4). It's not frame counts for games I'm talking about it, it's just sending data to VRAM then waiting 10mins for the CUDA cores to chew over it. Once the data is transferred, the PCIE speed doesn't materially affect the calculation of the image - that's down to the CUDA cores and other such things.

Anyway maybe 4x2 would see a hit more so than 4x4, but the question is, is it economically worth it to put the third 4090 on this board and lose 5%, or build an entire second machine with all associated costs (or build a threadripper machine, well that's certainly not economical at all, I'd take the 5% hit rather than spend another $3k on parts).
Even then it's only the 3rd slot taking the performance hit. The other 2 will render 3D as fast as anything else out there.

What I was more interested in was if the AMD board would see a 3rd slot speed of 4x2 with zero NVMEs installed, or how you tell which NVME slot to use that won't affect slot3's speed. But maybe it's better to use the intel board anyway, even if I won't be able to upgrade the CPU in 2yrs.

edit: It seems the AMD will run at x2 no matter what? Even if no NVME installed in that slot. This is what confuses me as I thought AMD had more lanes or at least the same as Intel, the way they "count" and market them is confusing. Someone else said Intel doesn't "count" chipset lanes which is why AMD is listed at 24 but Intel lists 20. But even so why the difference on that 3rd slot... I was simply hoping for "one simple trick" to run that 3rd slot at x4 not x2. :) Ah well.
You cant just add 4+2 and make 6, as 6x slots dont exist

And intel boards only do the lane swapping because they don't have enough lanes and they're all from the chipset - when the 5.0 NVME slots are from the AMD CPU, they cant mix with 4.0 ones from the chipset
 
You cant just add 4+2 and make 6, as 6x slots dont exist
Uh I didn't say that? 6?? Not sure what you read here. I said it's either going to run at PCIE4x2 or PCIE4x4, the manual makes it look like the slot will run at PCIE4x2 no matter what, even if the NVME slot is empty, which is dumb, but unsure if a BIOS update in the future could just see the slot run 4x4 when no NVME installed - depends on how it's wired? And the manual doesn't tell me that or if it does I didn't see it.
 
Back
Top