PCIE Gen 4 x8 and PCIE Gen 3 x16 *Not realistic* #ThoughtExperiment

Hardware GPU
1

Fair warning: This is rather long-winded till i get to the point, and its purely a thought experiment.

So I was thinking up a thought and wanted to get some opinions, and if anyone has a way to test this on hand i would appreciate that, if not then that’s totally fine, this is mostly just a thought experiment, and i couldn’t find any articles either here or on the rest of the interwebs crawled by Google.

The thought here is fairly simple, and i’m not super well versed in multi-GPU and Virtualization stuff.

Granted! In roughly a month or 3 this will be totally worthless, but its an interesting thought.

But lets Take an example of a Threadripper 2 CPU and Ryzen 3000 CPU both with exactly the same board, and it magically has both a PCIE Gen 3 and Gen 4 Layout. (Yes, i am fully aware that there are no boards with the capability of holding both types of very different chips)

All else being equal lets say you run Linux for some games that work exceptionally well on DXVK/Lutris/Proton or whatever you want to use in this experiment. And you virtualize windows for Adobe Suite, or some Niche games with silly DRM.

Threadripper can handle dual x16 graphics cards, however X570/Ryzen 3000 can only handle X16 and x8.

In addition to the CPU and Motherboard let’s take 2 perectly the same (including silicon) 5700 XTs.

First test would be the Threadripper Virtualizing Windows and both cards are running PCI 3.0 x16 speeds. Game 1 in Linux runs at a locked 60 FPS and has almost no Frame TIme Spikes. Game 2 perfectly virtualized in windows being identical runs at a locked 60 FPS and almost no frame time spikes.

Second Test would be the x570 Ryzen 3000 Chip with identical cores/threads/clocks ot the threadripper. The only difference would be it would run PCI Gen 4. Virtualizing windows with the linux Card Running at x16 and the Windows Virtual Card running at x8.
Assuming the Linux is exactly the same locked 60 FPS with no Frame Time Spikes…

Should the WIndows Client run almost identically with almost no difference? Thought being that Gen 3 x16 runs at 15760 MB/s and Gen 4 x8 runs at 15752 MB/s so a matter of a 0.0005% difference in performance. And I believe a Double bus speed of Gen 4.

2

I’m not sure why this is even a thought experiment to be honest.

The graphics cards you mention don’t benefit from pcie gen 4 as they can’t saturate pci gen 3. You also have the inherent architectural and chipset differences that come into play. There are only edge cases where one may be better than the other to a noticeable standard.

Really there are reasons why someone may choose a 3900X/3950X over a TR 2920/2950 and reason for the reverse. IOMMU grouping and USB/TB3 options being one of many.

1 Like