Would a 7900XTX be bottlenecked by a Pci-E 3.0 x8 port?

[Badstormer]

I was checking the hardware specs for the motherboard I'm using with this GPU and realized that it doesn't even support gen 4 Pci-E, and the highest speed I can get on it with the current configuration is Pci-E 3 x8. How badly might this limit a 7900XTX that would be used for things like high-end VR, video editing, or other demanding applications?

[C1REX]

引用自 Badstormer：

I was checking the hardware specs for the motherboard I'm using with this GPU and realized that it doesn't even support gen 4 Pci-E, and the highest speed I can get on it with the current configuration is Pci-E 3 x8. How badly might this limit a 7900XTX that would be used for things like high-end VR, video editing, or other demanding applications?

The difference is within margin of error for gaming. Even 4090 is not fast enough to take meaningful advantage of PCIe 4.0.

https://youtu.be/v2SuyiHs-O4?si=wD_2bM_zwbWBUC16

[nullable]

This is often a concern, not using the latest slot/or the fastest slot. But the thing to remember is you're often not going to be using 100% or close to it bandwidth on the current slot, or even the previous slot. PCI-E has done a decent job overall of keeping ahead of user/hardware needs.

PCI-E 4.0 wasn't released because the next gen hardware (at the time) was going to be crippled by PCI-E 3.0. Or because current gen hardware (at the time) had maxed out PCI-E 3.0.

People have been fretting over this for every version of PCI-E and every version of hardware that supports the newest version of PCI-E. And it's almost always the same story, "not really an issue". Maybe if you were to go back to PCI-E 1.0 or 2.0 you might be able to get some meaningful results. But part of the problem would also be the hardware platforms using those standards is so old now that would do as much to limit performance as anything. I mean using a modern GPU with a Core 2 Duo, even under optimal conditions is going to result in some bottlenecks.

[domigold]

The general rule is: As long as you have VRAM Ressources left on your GPU, your performance loss should be small or zero (below 5% i would guess). If your VRAM gets filled completly, then you lose a very big chunk of performance, since your GPU outsorces Data to your RAM. Then you lose performance, because first your RAM Speed isnt as fast as your VRAM Speed and second you dont have enough PCI Lanes to transfer data. That results in moderate to very heavy performance losses (depending on what exactly you are doing).

In Gaming you should be fine I think (no idea how performance heavy VR is these days). In Video Editing you should be running into problems very easily and lose a big amount of performance

[Badstormer]

引用自 domigold：

(no idea how performance heavy VR is these days). In Video Editing you should be running into problems very easily and lose a big amount of performance

In this case, I do both regularly, and VR is much more performance heavy than flatscreen gaming, especially on something like an Index.

It seems that for most people, Pci-E doesn't matter, but when it comes to much more GPU-demanding stuff, it does start to matter.

[Badstormer]

引用自 smallcat：

I hope you re not gonna use a Pentium processor !? The card in question is expensive , so your PC cant be so old . i guess it supports x16 pci-e 3.0 .What mobo do you have ? As for the loss , usually 5-10% and very rarely 20% .

I'm currently using a Ryzen 7 5800X, and the motherboard is a B-450F. The M.2 slots are both in use, which apparently limits Pci-E lane availability.

edit: I have an X570 lying around I could replace the b450-f with, and I'm open to input on whether this will be enough to avoid pcie bottlenecks with the m.2 slots in use. I believe the X570 uses pcie 4, but I'm not sure if the lanes will be limited if I utilize the m.2 slots.

[r.linder]

For x16 GPUs it doesn't really make a difference. For x8, it makes a difference of a few frames. For x4, it can make a pretty big difference to run less than the intended bandwidth, which is why cards like the 6500-XT and 6400 sucked.

[AbedsBrother]

引用自 Badstormer：

引用自 smallcat：

I hope you re not gonna use a Pentium processor !? The card in question is expensive , so your PC cant be so old . i guess it supports x16 pci-e 3.0 .What mobo do you have ? As for the loss , usually 5-10% and very rarely 20% .

I'm currently using a Ryzen 7 5800X, and the motherboard is a B-450F. The M.2 slots are both in use, which apparently limits Pci-E lane availability.

The speed hit comes on the sata ports afaik, not on the pcie slot speed, though that might vary from board to board (I have a 3700x on an Asrock B450m Pro 4).

[r.linder]

引用自 AbedsBrother：

引用自 Badstormer：

I'm currently using a Ryzen 7 5800X, and the motherboard is a B-450F. The M.2 slots are both in use, which apparently limits Pci-E lane availability.

The speed hit comes on the sata ports afaik, not on the pcie slot speed, though that might vary from board to board (I have a 3700x on an Asrock B450m Pro 4).

https://rog.asus.com/ca-en/motherboards/rog-strix/rog-strix-b450-f-gaming-model/spec/

引用自 ASUS：

When the M.2_1 Socket 3 is operating in SATA or PCIE mode, SATA6G_5/6 ports will be disabled.
When the M.2_2 is occupied by M.2 device, PCIe x16_1 will run at x8 mode.

The second slot shares bandwidth with the top PCI-e slot.

[Badstormer]

It is my understanding that the X570's primary expansion slot doesn't have to ration its lanes with SATA/M.2 slots like the B-450F does. Assuming I'm actually correct about this, it seems it may be time for me to swap out motherboards.

[r.linder]

引用自 Badstormer：

It is my understanding that the X570's primary expansion slot doesn't have to ration its lanes with SATA/M.2 slots like the B-450F does. Assuming I'm actually correct about this, it seems it may be time for me to swap out motherboards.

The bandwidth cut doesn't make any real difference, your CPU will bottleneck the 7900-XTX more than the slot ever will.

If you're going to change the motherboard, you may as well go up to AM5 and get a 7800X3D and DDR5-6000 CL30 RAM on top of that, because at least you'd get a tangible benefit. You won't really notice any difference at all if you just get an X570 motherboard. Literally the first comment to this thread shows this.

[Badstormer]

引用自 r.linder：

引用自 Badstormer：

It is my understanding that the X570's primary expansion slot doesn't have to ration its lanes with SATA/M.2 slots like the B-450F does. Assuming I'm actually correct about this, it seems it may be time for me to swap out motherboards.

The bandwidth cut doesn't make any real difference, your CPU will bottleneck the 7900-XTX more than the slot ever will.

If you're going to change the motherboard, you may as well go up to AM5 and get a 7800X3D and DDR5-6000 CL30 RAM on top of that, because at least you'd get a tangible benefit. You won't really notice any difference at all if you just get an X570 motherboard.

Does this still apply with things that are very GPU-oriented like hardware video rendering/neural network training? In my experience, the GPU is virtually always the bottleneck in these cases.

edit: also, the OP specifies use cases other than gaming, the first post in this thread only addresses gaming

[Illusion of Progress]

I use an X570 board with a PCI Express 4.0 GPU and a pair of PCI Express 4.0 SSDs. They all run at that.

I don't know if I'd buy an X570 board if you already have a B450 board, but if you really have one "laying around", that's an easy recommendation to change over to it here. I'm a bit confused why you wouldn't have been using it to begin with, but that's not my business and it doesn't matter. If you have it, it would alleviate your issues.

If it were just PCI Express 3.0, I'd say it's fine and maybe not even worth the effort unless you wanted to, but PCI Express 3.0 at half the lanes? That's PCI Express 2.0 at full lanes bandwidth. Even that might be still most likely fine outside edge cases, but yeah I'd switch it there (again, since you have the board "laying around").

[r.linder]

引用自 Badstormer：

引用自 r.linder：

The bandwidth cut doesn't make any real difference, your CPU will bottleneck the 7900-XTX more than the slot ever will.

If you're going to change the motherboard, you may as well go up to AM5 and get a 7800X3D and DDR5-6000 CL30 RAM on top of that, because at least you'd get a tangible benefit. You won't really notice any difference at all if you just get an X570 motherboard.

Does this still apply with things that are very GPU-oriented like hardware video rendering/neural network training? In my experience, the GPU is virtually always the bottleneck in these cases.

edit: also, the OP specifies use cases other than gaming, the first post in this thread only addresses gaming

Either way there’s zero point in buying a new motherboard for no difference, when the 7900xtx can be bottlenecked by Zen3

[C1REX]

Zen3 may or may not bottleneck 7900xtx.
Depending of games and resolution played.

[Illusion of Progress]

引用自 Badstormer：

edit: I have an X570 lying around I could replace the b450-f with...

I think people missed this.

It's laying around. There's therefore no reason to skip on swapping to it and buying a ton of new parts, all for a very incremental upgrade over what they have now.

At least, that's my thought process.

Everything bottlenecks. You're not going to remove that by moving to a one generation newer platform.