CPU Current Gen4/PCIe 4.0 GPU on x4

integer

Trash Compacter
Original poster
Aug 29, 2019
34
5
Anyone considered running a new PCIe 4.0 GPU on a x4 4.0 slot?

I'm on X570 so limited to 20 lanes. (Threadripper is great for lanes, but it's big, expensive, and slow on singlecore. so 5000 series + x570 it is)

Anyhow, the primary x16 4.0 slot seems like a bit of a waste for a 4.0 GPU equipped GPU.

This would be like similar data rate to if a x32 PCIE 3.0 slot existed, which we can safely say is overkill for gaming scenarios. YMMV (eg maybe machine learning its useful?)

Hence wanting to run the GPU on the remaining x4 (x16 physical) 4.0 slot (Slot 3 in my case).

In theory, when paired with a 4.0 GPU (Nvidia RTX 3000 series / AMD RX 6000 series) it should equate to a 8x PCIE 3.0 data rate. Which when tested only suffers a small (few percent) fps decrease over x16 3.0. At least that's what I think.

Maybe this gives someone else some ideas to about new build options. I could also be wrong, so just happy with any feedback. Even if it's to say "only an idiot would do that" ?
 
Last edited:

Runamok81

Runner of Moks
Jul 27, 2015
445
621
troywitthoeft.com
There is surprisingly little information about measuring the performance hit of dropping PCIe 4.0 x16 down to x4. Theoretically, it should equal PCIe 3.0 x8, and that HAS been tested and - as you stated - should get you 99% performance. My biggest question here is, how do you intend to test? Can you explain?

Because we now have NUCs with PCIe 4.0, I could see the test results being of interest to the folks that are currently leveraging their PCIe 3.0 M.2 x4 slots to attach eGPUs, such as the @REVOCCASES example here. The biggest hurdle for that crowd, is that we are still lacking an M.2 to PCIe adapter for 4.0, yes?



Regardless, a PCIe 4.0 x4 test results are useful. They would provide valuable information to those folks waiting on the adapter. And, if anyone has seen an M.2 to PCIe 4.0 adapter, then that would really make things interesting! Good luck.

Update: I did find a PCIe 4.0 riser ADT-Link G402-3, but no M.2 adapter, yet.
Super risky, but someone could also try their PCIe 3.0 M.2 adapter over 4.0 and hope for the best.



But that seems risky. I believe the electrical signals of 3.0 and 4.0 are very different. I like it.
 
Last edited:
  • Like
Reactions: integer

Quango

SFF Lingo Aficionado
Apr 6, 2019
102
34
A future SFF candidate:

When Rocket Lake arrives, the Desmini H470M-STX mainboard will have a functional PCIe 4x Gen4 "Hyper M.2" on the backside.

 

integer

Trash Compacter
Original poster
Aug 29, 2019
34
5
There is surprisingly little information about measuring the performance hit of dropping PCIe 4.0 x16 down to x4. Theoretically, it should equal PCIe 3.0 x8, and that HAS been tested and - as you stated - should get you 99% performance. My biggest question here is, how do you intend to test? Can you explain?

My answer may disappoint cos I'm kinda doing the whole SFF thing on easy using an Asus x570 mATX board. 😂







Because we now have NUCs with PCIe 4.0, I could see the test results being of interest to the folks that are currently leveraging their PCIe 3.0 M.2 x4 slots to attach eGPUs, such as the

Interesting didn't know that . Thanks
 
  • Like
Reactions: Runamok81

Kommando

Average Stuffer
Dec 19, 2020
78
54
If you look at this comparison you see that even PCIe 1.0x16 wasn't that bad compared to 3.0 when it was released.

Sure these are way older cards, but PCIe 4.0x4 equals PCIe 2.0x16 already.

Additionally, most external "homebrewn" graphics cards use connections like PCIe 2.0x4 or similar. It's all not that punishing, unless the card needs to transfer a lot of data, i.e. VRAM isn't big enough.
 
  • Like
Reactions: integer

integer

Trash Compacter
Original poster
Aug 29, 2019
34
5
So I moved the RTX 3080 to the x4 PCIe 4.0 slot. And ... performance was the same as the x16 slot! Couldnt believe it.

Scored 16800 in Timespy. So looks like I'll keep it there and have the x16 available for Quad NVME :)
 
  • Like
Reactions: Runamok81

Runamok81

Runner of Moks
Jul 27, 2015
445
621
troywitthoeft.com
I mean, it's not THAT unbelievable, right? 😜
Sure, nobody has tested it, but all signs point to it working this way. Thank you for confirming.

TechPowerUp has a good read on PCIe 4.0 scaling here.



There is probably some performance difference, but not much. According to the article, even with an RTX 3080 we're still not even close to saturating PCIe 3.0! Once again though, that article did NOT test PCIe 4.0 x4 directly, so you are in a good position to run a round of tests to see if there truly is a difference or not. Good to hear your tests are looking good. There are some folks with x4 PCIe M.2 slots that might want to try hooking up a GPU.
 
Last edited:
  • Like
Reactions: Phuncz

tinyitx

Shrink Ray Wielder
Jan 25, 2018
2,279
2,338
If we approximate PCIe 4.0 x4 =~ PCIe 3.0 x8 =~ PCIe 2.0 x16, then TPU's graphs show the difference between PCIe 4.0 x4 and x16 is 4%.
When one is set on setting a synthetic benchmark record, then 4% is definitely considerable and measurable.
But if one is a gamer and not a benchmarker, then 4% in framerate is visually indistinguishable. I do not think a normal person can tell 96 fps vs 100 fps.
 
Last edited:
  • Like
Reactions: Phuncz and rfarmer