Asus Z690 Hero + RTX 5090 requires PCI Express Gen.3 / 4 for stability

[darkcg]

Hello, I'm experiencing a weird situation with my Asus Z690 Hero Motherboard and my new RTX 5090 card.
The system goes black screen as soon as it reaches the desktop (sometimes even before, when loading the operating system). Basically it is unstable and it won't allow you to make any work at all. This happens not only in Windows but also in Linux (Ubuntu 22.04 LTS) so there is nothing specific to the operating system, I think.

But I discovered that putting the PCI Express from "Auto" to Gen.3 fixes the issue, I've put the system under stress tests using 3DMark and everything seems to be fine. The only problem is I'm losing a good 10% of performance with this setting alone.

Any suggestion? Do you think it's a problem with the motherboard BIOS?

EDIT: Of course I have the latest BIOS version (4101) and the latest nvidia hotfix driver (572.75), both on Linux and Windows.
The card is a Zotac RTX 5090 AMP Extreme Infinity. I can't do other tests with other models right now, but a new Gainward RTX 5090 Phantom is arriving, so I'll do other tests.

[ElectroStingz]

Hello,

This is an issue with the 5000 series relating to PCI-E 5.0.

https://www.guru3d.com/story/troubleshooting-rtx-5090-black-screen-failures-switch-to-pcie-gen-40/

https://www.pcgamer.com/hardware/graphics-cards/nvidia-says-it-really-has-sorted-rtx-50-series-black...

You should contact Zotac and see if they have a beta BIOS update for the card (I couldnt see anything on their site)

[darkcg]

Thanks for the answer but I'm curious. What makes you think that it's the card's manufacturer that must fix the issue rather than the motherboard's manufacturer?
Also, switching to Gen.4 doersn't work for me, I must switch to Gen.3 so it doesn't seem to be the same issue. I refuse to think that the Gen.4 is "too fast" for the initialization of this card to be honest. 🙂 And Gen.4 was working with my previous RTX 4090.

The second link you've provided is a completely different issue

[ElectroStingz]

Why not the motherboard? Because for the most part it has been widely accepted that the way nvidia have designed the card is likely causing a lot signal integrity issues for the PCI-E interface. Plus depending on other factors, EMI, card design, PSU, voltage supply (how clean the voltage line is, quality of VRM and noise suppression)... your overall interferrence could be a lot worse.

The motherboard slot and design will also impact the signal quality but then why does it work fine with your previous 4090 card? So is it a compatibility issue or a signal issue, Gen 4, Gen 3 both decrease frequency which helps with PCI-E signal quality. So if I were to pick which one is the main problem I would go with the Zotac card.

But equally one might take your Zotac card, plug it into another PC and it works fine without issues due to the differences mentioned above. So what's the cause? 🙂

Hopefully the Gainward card works better.

[darkcg]

Cool, but "accepted" by whom? Users? Has nVidia acknowledged the issue? many people think it's a motherboard issue.

We had gen5 motherboards for 3 years on the market but without any cards to test them with, there were only SSDs. The quality of the implementation and signal integrity on each board is going to vary and I don't think firmwares will fix it. I suspect that the majority which claim to have PCIe 5.0 support can't actually achieve the signal integrity required for real world use (especially for all 16 lanes). For some boards it may be resolved with BIOS updates that will update the link training routines but for many I suspect that all they'll do is effectively cap the speeds at PCIe 4.0 or at least below the maximum PCIe 5.0 speeds. 

If this is the case, I would like to know from ASUS. Because at this point this is a Gen.5 motherboard that can't handle Gen.5 peripherals.

[robfrawley]

---

@ElectroStingz wrote:

[...] for the most part it has been widely accepted that the way nvidia have designed the card is likely causing a lot signal integrity issues for the PCI-E interface.

---

A few important clarifications. I don't think it is clear to anyone what is causing the PCI-E signal integrity issues with the 50-series cards. Your comments about signal integrity concerns only apply to the Founder Edition, as it uses separate PCBs connected using a proprietary cable. However, this isn't true with AIB cards from ASUS, MSI, Zotac, or anyone else.

I have the issue described by the OP (display going black when Windows boots or soon after when using PCI-E Gen 5) using an MSI RTX 5090 Ventus 3X OC. Switching the PCI-E slot from Gen 5 to Gen 4 solves these issues for me (unlike the OP, who needed to downgrade to Gen 3 to achieve a stable system).

---

@ElectroStingz wrote:

[...] why does it work fine with your previous 4090 card?

---

The 40-series cards are PCI-E Gen 4 devices, a huge difference from NVIDIA's latest. The 50-series GPUs will be the first PCI-E Gen 5 device many people install in their motherboard's PCI-E slot. It was for me. Like the OP, I had no issues with my prior 4090 (but, again, that was a PCI-E Gen 4 device).

I have no idea if the issue is with NVIDIA or ASUS or whether it is software or hardware, and nothing that was said here leads me to believe one is more probable than the other.

[ElectroStingz]

I agree there is no actual confirmation on the issues yet however looking at the cards in question, founders or not they all seem to struggle with Gen 5 mode.

The founders edition card and detachable cable is actually part of the PCI-SIG PCIe specification known as CopperLink, provided this is within the design specification to the Gen5 spec then it's actually a better method. This design takes the signal straight from the GPU, avoids the PCB and any direct signal losses, interference or other and as it's cabling can be thicker (than PCB traces) also better for signal transfer. (If done incorrectly then of course it's not going to work well)

Also Gen5 Compliance testing is performed on all hardware running to this spec, motherboard and GPU manufactures have to comply to meet the spec. When you look at the issues it is not a pandemic and every single card failing but a portion so it's likely a combination of factors at play so it would be interesting to see how the OP's new Gainward card performs.

And my comment on why the previous card work was more to say, if the previous card worked in Gen 4 then it would suggest the motherboard is capable of Gen4, as the OP mentions only Gen3 working on the newer card.

[Silent_Scone]

Hi @darkcg Have you tried the hotfix driver? https://nvidia.custhelp.com/app/answers/detail/a_id/5640/~/geforce-hotfix-display-driver-version-572...

EDIT: I see that you have, apologies. This issue has been acknowledged by NVIDIA so hopefully they'll get to the bottom of it. 4000 series GPUs don't have a problem at GEN4 spec.

No issues here on X870E @ GEN5 with the Astral 5090 OC, and I'm using a LINKUP GEN5 riser cable @ GEN5.

[darkcg]

@Silent_Scone That's not completely true, even the 4000 series suffered this issue but the occurences were so sporadic that the problem has never reached the masses. My brother had this same issue 2.5 years ago using one MSI RTX 4090 Suprim paired with an Asus Z690 Extreme. In the end he solved the problem by changing the CPU (moving from a 12900K to a 13900K). nVidia at the times never acknowledged the problem and MSI didn't know what to say. We've never understood why changing the CPU fixed the problem. These kind of problems are scary.

[SneakyLittleman]

Thank GOD I came across your thread man! I was getting crazy! got a new Aorux 5090 yesterday and it kept crashing whatever I did - new cables / new PSU / old PSU... I just switched to PCIE4 in my Z690 Hero MB Bios and now it seems to be running fine... Hope this gets fixed, but at least this 5090 is useable. Cheers. (For me, on latest bios, no need to go down to pci3, I'll experiment & see if it's stable).