NVIDIA GeForce RTX 5090 PCI-Express Scaling

[W1zzard]

The new NVIDIA GeForce RTX 5090 is the first high-end graphics card that makes use of a PCI-Express 5.0 bus interface. Are you in trouble when try to run it on PCIe 4.0? What about x8, like when an SSD is using up some bandwidth? We've also tested various PCI-Express 3.0, 2.0 and 1.1 configs to get a feel for how FPS scales with bandwidth.

Show full review

 

[Chaitanya]

Any plans for CPU bottleneck testing?

 

[Readlight]

What is fastest PCIe 1.1 CPU Athlon 6000+

 

[Daven]

An external GPU enclosure using TB5 (similar to PCIe 5.0 x4) would give you almost all the performance. That's pretty nice.

 

[redeye]

wow, want the card but… finally answers my question… since scaling from pcie4x8 to pcie4x16 is not more than 2 fps on cp2077 on 5090, i not missing out on more frame2 on my 7900xtx because it was damaged and is stuck at pcie4x8.

(good news from a card that is the fastest… 5090)
RT is already lacking (vs nvidia) on the 7900xtx, so moot point for the scaling on a 5090 on ultra RT.

why not review this card at 8k, just so we can shutup those that want 8k on a console. HFW 4k120 or 8k30 (Guessing)

 

[Panchovix]

Many, many thanks for this! I plan to get a 5090 (or some 5090s) alongside other GPUs, and I could run it only at X8 5.0 or X4 5.0. The perf regression with those 2 are pretty low, so pretty nice!

 

[Dristun]

Thanks for this. Any plans for power scaling tests or too busy already? Like 4090 vs 5090 @300W, for example.

 

[sethmatrix7]

2 new reviews at once?!? W1zz.. I'm at work dude!

 

[IMABEARLOL]

I would love to see motherboards that only provide 5.0 x8 (essentially 4.0 x16) bandwidth to the GPU and reallocate the other 5.0 x8 lanes to M.2 and/or additional PCIe slots. I'd rather Give up 1-2% performance for much more expandability.

 

[Baba]

I was planning on building a new AM5 system along with a 5080 but might just keep my Taichi x370 PCIe 3.0 mobo and get a 5090 with the savings of not buying a new mobo and RAM. I play at 4k. What do you guys think?

 

[bug]

As usual, PCIe bandwidth is largely irrelevant. That's always a plus.

 

[wNotyarD]

I would love to see motherboards that only provide 5.0 x8 (essentially 4.0 x16) bandwidth to the GPU and reallocate the other 5.0 x8 lanes to M.2 and/or additional PCIe slots. I'd rather Give up 1-2% performance for much more expandability.

Huh? I thought most of Z790 boards cut the 5.0 PEG slot from x16 to x8 when installing an SSD to the 5.0 M.2 slot.

 

[Vagotron Prime]

I would love to see motherboards that only provide 5.0 x8 (essentially 4.0 x16) bandwidth to the GPU and reallocate the other 5.0 x8 lanes to M.2 and/or additional PCIe slots. I'd rather Give up 1-2% performance for much more expandability.

This is how it is already done. Take AMD's AM4 and AM5 platforms for example. These have 24 PCIe lanes, divided as 4 lanes to motherboard, 4 lanes to NVME, and 16 lanes to PCIe. Most motherboards will have more than a single NVME slot. One will use the dedicated x4 lanes, the others will "steal" lanes from the PCIe, or piggyback on the Mobo lanes.

 

[Voodoo Rufus]

It's surprising how well x16 2.0 holds up at 4K, but one wouldn't be likely to maintain that with a CPU from that era I think. x16 3.0 still holds up great.

Would love to see some CPU scaling from different eras. I would guess a Haswell can still hack it with some strain, but Sandy Bridge and older would really suffer from 2.0 PCIE. Would be fun to see how strong Coffee Lake still is.

 

[csendesmark]

It's surprising how well x16 2.0 holds up at 4K, but one wouldn't be likely to maintain that with a CPU from that era I think. x16 3.0 still holds up great.

Would love to see some CPU scaling from different eras. I would guess a Haswell can still hack it with some strain, but Sandy Bridge and older would really suffer from 2.0 PCIE. Would be fun to see how strong Coffee Lake still is.

Since 32GB VRAM so big, can hold some games whole. So there is not much need to talk to the mainboard

 

[InVasMani]

On a older system CPU/Memory bottlenecks would be the only practical concern.

 

[Auxityne]

I was looking forward to this to see how annoyed I should be with the way X870E does PCI-E lane sharing due to the... I'll call it interesting mandate of USB4. Seems like it doesn't really matter and I can just buy whatever board I want.

 

[lexluthermiester]

So to quickly sum up, PCIe bus versions are not a limiting factor to GPU performance. W1z has shown this several times in the past and it's still holding true. Not really surprising to anyone who understands how the tech works, but I digress..

 

[W1zzard]

Any plans for power scaling tests or too busy already?

Too busy, 5090 custom, then more releases from NV, and maybe AMD manages to launch something, too

Also I want to upgrade the CPU testing rig to 5090

 

[lexluthermiester]

Also I want to upgrade the CPU testing rig to 5090

Yeah, that would only be logical.

 

[ToTTenTranz]

Turns out if the driver can stick almost the whole game into the VRAM, the CPU doesn't need to send data to the GPU that often.

 

[lexluthermiester]

Turns out if the driver can stick almost the whole game into the VRAM, the CPU doesn't need to send data to the GPU that often.

The CPU is still processing much of the game code. GPU's are processing graphics data.

 

[Cipher908]

Awesome to know that the 5090 will still perform great with my Core2duo E8500 on PCIe 1, can't be having any bottlenecks )

 

[Imploser]

How to test lower PCIEX generations with same motherboard? I couldn't read if it was already written.

 

[lexluthermiester]

Awesome to know that the 5090 will still perform great with my Core2duo E8500 on PCIe 1, can't be having any bottlenecks )

You're failing at context. The bottleneck on that situation is more the CPU than the PCIe bus. The point of W1zzards efforts here is to assure the buying public that if they still have a board with PCIe 4.0 the card performance will not be adversely affected. Even an older gen board with PCIe 3.0 will not bottleneck this card by enough to matter.