4090 Voltage sag

[cakyo0120]

First of all, I have an Aero OC 4090 that had a burnt connector, which was replaced by Gigabyte back in August 2023.

The voltage readings in HWInfo—+12V PCIe, 16-pin VHPWR, and GPU FBVDD—were all fine, sitting around 12.230V at idle and 12.120V–12.160V under load. These readings remained consistent throughout 2023, 2024, and the first half of 2025.

However, a couple of months ago, those voltage readings started dropping below 12V, hovering around 11.8V–11.9V. I know that ±5% is considered "safe", but even when undervolted to 0.900V @ 2325MHz, the voltages wouldn’t stay above 12V. Given that my typical and consistent readings were around 12.120V–12.160V under load for many years, this sudden drop is very concerning to me.

 

At first, I suspected my Seasonic Prime GX-1000 was failing, so I bought an brand new NZXT C1000 (ATX 3.1). However, it performed worse than my 3 year old GX-1000, with voltages dropping slightly lower.

After testing and ruling out the PSU as the cause, I RMA the card. Gigabyte told me the card was working fine with NFF (No Fault Found), and they said they tested it using Nvidia’s diagnostic tool.

 

The issue is, I’m not sure why the voltages suddenly dropped a couple of months ago. Even though the readings are "technically" within spec, my typical voltages were consistently well above 12V—undervolted or not.

Does anyone have a clue what could be degrading?

 

I've also inspected my motherboard, particularly the VRM area, and there are no burn marks or bulging components whatsoever. I also switched the PCIe from Gen 4 to Gen 3, but the voltage readings remained the same. Therefore, I'm not sure if the issue is related to the PCIe slot.

Specs:

• AORUS Z690 Pro
• Intel i7-12700K
• RTX 4090
• Trident Z5 4×16GB
• Seasonic Prime GX-1000

By the way, all cables are original PSU-provided cables. For the 16-pin, I’ve tried both 12VHPWR to 2×8 and 12V-2×6 to 2×8 configurations the results were the same.

 

As for those screenshots, I stopped the test before taking them. That's why the "Current" reading looks fine on some of the screenshot, but the "Min" value reflects what was happening when the test first started.

GX-1000-OCCT70% RUN

 

GX-1000-0.900v@2620

 

 

C1000ATX3.1-OCCT100%

 

 

C1000ATX3.1-OCCT50%

(As you can see i only start running for 39sec and the voltage already dropping)

 

[leclod]

Couldn't the "drop" result from an update, be it from HWinfo, Bios, NVidia ... ?

One of your headers on psu side got a little looser or corroded and resistance slightly increased ?

So today you have no problem at all ?

I'd call it a (good) day.

(In case you're worried about melting the connector again, you could survey it's and the cable's temperature with an infrared gun or with an external thermal sensor. Both can be found cheap.)

Edited August 8, 2025 by leclod

[cakyo0120]

12 hours ago, leclod said:

Couldn't the "drop" result from an update, be it from HWinfo, Bios, NVidia ... ?

One of your headers on psu side got a little looser or corroded and resistance slightly increased ?

So today you have no problem at all ?

I'd call it a (good) day.

(In case you're worried about melting the connector again, you could survey it's and the cable's temperature with an infrared gun or with an external thermal sensor. Both can be found cheap.)

Unfortunately, no, the problem still hasn't been fixed.

 

Ever since the connector melted, I've been extra cautious about my GPU's behavior. The voltage readings don't seem low enough to have caused the melt, but it's strange how they suddenly started dropping to levels I've never seen in years. (The previous melt was due to the first-generation CableMod angled adapter, but I no longer use any extensions or adapters.)

 

I've reseated and checked every cable to make sure everything is perfectly and firmly plugged in, but the issue persists.

I've tried everything you could think of like DDU, Windows reinstall, BIOS updates (both motherboard and GPU), and even BIOS rollbacks and even take every component out the case and build it like a fresh PC

That's why I'm so lost here.

 

One thing i notice is that the GPU FBVDD reading would drop to 11.900v even at idle but just for couple seconds, which it never happen on my system before