[Jaquith]

My copy of GPU-Z (version 2.51.0) reports Bus Interface as "PCIe 4.0 @ 8x" on my PCIe 5.0 Slot / PCIe 4.0 GPU. My ASUS ROG Extreme Z790 has a PCIe 5.0 x16 Bus Interface and my RTX 4090 has a PCIe 4.0 x16 Bus Interface, I'm crystal there. Reporting confusion .. is GPU-Z picking up on the "x8" on the PCIe 5.0 side and reporting it as PCIe 4.0 "x8" and if so what is the actual bandwidth to the RTX 4090?

Yesterday I watched a video posted by Der8auer on YouTube entitled "I tried to repair my RTX 4090 and killed it " where we are using the same MOBO using different RTX 4090s and both of us have the same reporting of "PCIe 4.0 @ 8x." Also obviously I double checked the "shared" limits for my MOBO. I'm using the DDR4 DIMM.2_SLOT with 2x M.2 drives and the M.2_1(SOCKET3) with an M.2 all M.2's are PCIe 4.0 types. Manual: M.2_1 shares bandwidth with PCIEX16(G5)_2. When M.2_1 is enabled, PCIEX16(G5)_2 will be disabled. The RTX 4090 is installed in the PICEX16(G5)_1.

=> While typing this I have to see if there's a PCIe 4 .0 (G4) / PCIe 5.0 (G4) BIOS selection. Unfortunately the manual doesn't list BIOS entries to search handily.
=> Reporting bug or what?
=> What is my actual bandwidth to the RTX 4090?

So my question is has anyone with a Z790 MOBO experienced the same problem? My hope is the RTX 4090 is getting x8 of PCIe 5.0 bandwidth and GPU-Z can't report it properly. However, x8 of PCIe 4.0 bandwidth to my RTX 4090 is problem if accurate.

GPU-Z TechPowerUp
CPU-Z Intel Core i9 13900K @ 5501.32 MHz - CPU-Z VALIDATOR

 

[Jaquith]

You probably have something else plugged into the motherboard stealing lanes from the GPU. It could also be a bad mount of the CPU, missing pin(s) causing dropped PCIe lanes.

It is NOT the GPU getting x8 of PCIe 5.0 bandwidth. It cannot work that way, the 4090 does not support PCIe 5.0 and cannot use PCIe 5.0 lanes at all (it would get 0 MB/s bandwidth for a PCIe 5.0 lane). Each PCIe lane is its own wire with very different signals running on them, it is not something abstract that could combine/split up like that.

I seriously doubt it's CPU pins (I'd bet serious money) and if it were (I had a stroke) I'd probably get, almost 100%, a host of other POST ERROR CODES and/or POST FAILURES. LOL not my first PC. Der8auer and I both have the SAME EXACT ASUS Z790 Extreme but different RTX 4090's AND SAME PCIe 4.0 x8 per GPU-Z. As far as "stealing" bandwidth I have x2 USB 3.2 Gen 1 headers connected and x1 USB 3.2 Gen 2 (USB-C) connected. That's it. I can RTFM.

So it's either a GPU-Z Bug or ==>> ASUS Extreme BIOS BUG. I'm probably going to try it in my old PC, but what's the point. I'm going on statistical probability.

Old PC:
GPU-Z TechPowerUp
CPU-Z Intel Core i7 12700K @ 4700 MHz - CPU-Z VALIDATOR

 

[Jaquith]

Der8auer had the same issue with his Strix. Tried to fix it with a heat gun and crippled it further.

He ASSUMED the problem was from the GPU. Since we have the same ASUS Extreme Z790, it's probably the ASUS BIOS.

 

[Jaquith]

To confirm, do you have a M.2 drive in the first slot?

I have a total of three (3) M.2's installed. all are PCIe 4.0 types. A Samsung installed in the M.2_1 (SOCKET3) and Crucial two (2) in the DDR4 DIMM.2_SLOT. The M.2_1 (SOCKET3) is a PCIe 5 slot which is located above (under the OLED) the GPU PCIEX16(G5)_1 slot. The reason I chose all of those slots is so I can have access once I install all the EK blocks .. i.e. easy access!

 

[Jaquith]

So I would suggest to again RTFM like you said you know how to do.

M.2_1 shares bandwidth with PCIEX16(G5)_2. When M.2_1 is enabled,
PCIEX16(G5)_2 will be disabled

Since you have a M.2 drive in that slot, you are getting halved lanes because it shares GEN 5 lanes.
It does not matter that you have a gen 4 drive.

So the issue is not a bios bug or a gpu-z bug, only user error lol

LMAO .. only one problem with your theory the GPU is installed in as I said "The RTX 4090 is installed in the PICEX16(G5)_1. 1" .. not in PCIEX16(G5)_2.

 

[Jaquith]

Yeah, then why did his other Palit work just fine on the same mobo while his crippled Strix ran at 8x on two different mobo's? The problem was the GPU and it's not an uncommon occurrence.

Only one problem with your theory, go to 10 minutes in it's a DIFFERENT MOBO!

 

[Jaquith]

Please read carefully.

PCIEX16(G5)_2 AND M.2_1 SHARE BANDWIDTH WITH PCIEX16(G5)_1

You cannot use EITHER of those if you want PCIEX16(G5)_1 to have full x16 bandwidth.

WOW .. NO! Here's a copy of the page and below that at 10 minutes using a different MOBO.

 

[Jaquith]

Are you purposely trying to pull out rebuttals? der8auer's issue and your issue isn't the same thing.

He has a faulty GPU and you have reading comprehension issues.

//

Answer is it's not possible so that's why the board internally switches the GPU to 8x when it detects either M.2_1 or 16x Gen5_2 slot is used.
16 + 4 != 16

Der8auer damaged his GPU by heating it up in an effort for fix a "problem' that I don't think he had. Burn this into your head .. my GPU is installed in the slot PICEX16(G5)_1 not (_2) which is the bottom PCIe x16 length slot. PICEX16(G5)_2 is Disabled. The shared bandwidth of the Samsung in M.2_1 only effects PICEX16(G5)_2.

Look and re-read my original post and then look at the page of the manual and if necessary the actual picture I posted of my PC to see for yourself.

 

[Jaquith]

It doesn't matter what GPU slot you are using if you have an SSD using the Gen 5 capable M2 slot regardless of motherboard. Intel has no dedicated Gen 5 lanes for an SSD. If you are using the purely Gen 4 M2 slots (which are usually chipset on Z790) then you won't have this happen.

It's a definite possibility that it's not properly documented. IF ASUS is using the M.2_1(SOCKET3) effectively as "Dual VGA" then yep x8/x8 PCIe and can only utilize a "PCIe lane" regardless of its actual bandwidth. The test would be to move the M.2_1(SOCKET3) to e.g. M.2_2(SOCKET3) or M.2_3(SOCKET3).

Per Intel https://www.intel.com/content/www/u...-36m-cache-up-to-5-80-ghz/specifications.html
PCI Express Revision
5.0 and 4.0
PCI Express Configurations ‡
Up to 1x16+4, 2x8+4
Max # of PCI Express Lanes
20

‡ This feature may not be available on all computing systems. Please check with the system vendor to determine if your system delivers this feature, or reference the system specifications (motherboard, processor, chipset, power supply, HDD, graphics controller, memory, BIOS, drivers, virtual machine monitor-VMM, platform software, and/or operating system) for feature compatibility. Functionality, performance, and other benefits of this feature may vary depending on system configuration.

 

[Jaquith]

Never use the Gen 5 slot because it will use lanes from your PCIEX16 slot regardless if your M2 is GEN 4. T

Well "never" I wouldn't go that far. The problem is accessibility especially once you add water blocks. The M.2_1(SOCKET3) and DIMM.2 locations are easy access for upgrades and or failures without draining a loop. It's a reasonable assumption and a poorly integrated once not to mention poorly documented.

 

[Jaquith]

Yea I get what you're saying, I would say poor implementation too. Even when GEN 5 drives are available would it be worth to to sacrifice the GPU lanes for minimal gains? Not gonna start with GEN 5 GPU's because that would render than slot void. Oh well.

The only reason I used that top M.2 slot was for accessibility (just in case). However, as far as "worth it" .. for my uses cases at home absolutely not, I couldn't tell (feel / sense) any difference when I upgraded from a Samsung 970 EVO to the Samsung 980 Pro. I 'upgraded' in the sense my step daughter got the EVO in her Dell Notebook and the 980 was on sale. Now if I were running SQL tests at home then yep, I use to use a RAM drive.

Right now I'm just making certain everything is working and using the time to burn-in the new rig until the 4090 blocks come arrive from EK. Guessing there's a fair chance that I'll use the M.2_2 socket and if it dies I'll just disable it and install a replacement in the M.2_1 socket until I would drain and clean.

Anyway cheers! I'll post the GPU-Z after the rebuild it, I'm absolutely going to test.

 

[Jaquith]

How much performance is lost on a 4090 when running PCIe 4.0 x8?

Short answer 0-10 FPS depending upon the game and resolution. I'll run some benchmarks before making changes too difficult.

My use case is sub-4K but high FPS. LG 45GR95QE or Corsair Flex 45WQHD240 which are 240Hz @ 3440x1440 OLEDs.

 

[Jaquith]

I don't know if I'm tripping or not but OP's gpu-z image shows that it can max pcie 4.0 x16 but that it is in x8 1.1, which is essentially low power mode. Give me a image of gpu-z under full load and then you can say its not reporting correctly but so far it looks correct.

It was under no load (desktop only), the lanes x8 vs x16 is what I was looking to fix.

I have no issues pegging the 300 FPS max in Call of Duty @ 2560x1400 on my old monitor .. and if (edit) I never ran GPU-Z I wouldn't have noticed a thing.

 

[Jaquith]

You're running at 4.0 x8 as others have pointed out already. You need to avoid using the Gen5 M2 slots because the CPU only has 16 G5 lanes

Just plug your M2 into a Gen4 slot instead, then it will be using the Gen4 lanes on the chipset instead of the CPU's Gen5 lanes

With a 4090 you might actually see a small difference between x8 and x16 in that use case. I'm using a 3080 to run 21:9 1440p at 144hz, so I choose to run a 980 Pro on the Gen5 CPU lanes and run the 3080 at Gen4 x8.

1. According to the manual it should work at PCIe x16. The manual was poorly translated, you have to assume foot connected to the leg the leg connected to the hip and so on. Der8auer (Roman) read it the same way I did... To be clear I haven't tested anything yet to confirm or deny the solution. Right now I understand the logic. On the Z690 also with three M.2's it absolutely has no effect on PCIe x16. So if the Z790 chipset is more less the "same" I shouldn't have an issue. There is no conservation of bandwidth just lanes.
2. If you read any of my replies then you would understand WHY I chose the M.2_1 socket, the others are UNASSESSABLE once I install the EK blocks.
3. My 'old' rig uses an RTX 3090 with a 175Hz 3440x1440 and without DLSS there's no way I'd kill-off any FPS unless I had no other choices.

 

[Jaquith]

As others have said, the manual (you even posted the page with the details) is quite clear.

This scenario is not at all uncommon on consumer boards, where the CPUs simply don't have the PCI-E lanes of the workstation/server chips

Like what MOBOs? Got a list or even one example other than this one which has not been substantiated. Apparently Roman doesn't agree. Also where exactly does it say in the manual using M.2_1 makes PCIEX16(G5)_1 PCIe x8?

Inaccessible = cannot get to.
Unassessable = cannot evaluate.

Definitions of unaccessible

1. adjective
   capable of being reached only with great difficulty or not at all.

In the US favour is spelled favor.

I get a couple of you TROLL like little children and somehow feel (LMAO) empowered. But you're both (Agent-A01) guessing and assuming, and if you were in my home / office I'd have you escorted to the curb. So go Troll someone else. I don't need or want your "help."

 

[Jaquith]

Here we have a great example of someone who thinks they are smarter than everyone else.

Both of us have been here for over a decade yet we are here to troll some kid because why else would we be posting here.

@ every else, just ignore this dude and move on. A new bonehead to ignore

Then you seriously need to get a life. I read some of your other posts, you got a real passive-aggressive thing going on. Tell Roman he's "dense." BTW post your CPU-Z .. I got a feeling here about you.