I bought a second-hand Gigabyte AORUS RTX 5090 about a month ago. It works normally in light use, but demanding games such as Doom: The Dark Ages eventually cause the display signal to shut off. The rest of the computer stays running: game audio continues, peripherals remain connected, and the GPU fans keep spinning, but the display does not return until I hard-reset the system.
OCCT found no VRAM errors, and the problem happens on both CachyOS and a fresh Windows 11 installation, so it does not appear to be specific to one operating system or driver. Monitoring with HWiNFO and GPU-Z showed what appears to be a very large core-to-hotspot temperature difference. For example, the reported GPU temperature was around 66°C while the hotspot reached approximately 107.4°C, shortly before the signal disappeared. Limiting the card to 70% power in MSI Afterburner prevents the crash and greatly reduces the reported temperature difference.
This makes me suspect a poor thermal interface or uneven cooler contact, but I am not experienced with GPU repairs. I am also unsure whether the 5090 exposes a genuine hotspot sensor through these monitoring tools or whether the reading is being interpreted incorrectly.
The card is still covered by the second-hand store's warranty. The store is a general resale shop rather than a PC hardware specialist, so they cannot repaste it for me. Should I return it or request a refund, have a qualified third party inspect it, or attempt a repaste myself? I am concerned that opening the card could damage it or invalidate the warranty, especially given its value.
4 Answers
Repasting a graphics card is possible, but a high-end card like this is expensive and easy to damage. Removing the cooler can disturb thermal pads, damage small components, or leave mounting pressure uneven. I would only consider it after confirming that the warranty is unusable and, ideally, having a repair technician do it. Returning the card is less risky than experimenting on it yourself.
The card may be hitting a protection limit or becoming unstable when its full power and boost behavior are used. Check the power supply, GPU power connectors, and cable seating as well, since a marginal power connection can also cause a display failure while the rest of the system continues running. However, because lowering the power limit consistently fixes the issue and the card is under warranty, documenting the temperatures, settings, and crash behavior and then making a warranty claim is still the best course.
Since the card is still under warranty, the safest option is to return it or have the seller authorize a repair. A large temperature difference can indicate poor cooler contact, dried or uneven thermal paste, or an issue with the cooler mounting, but opening the card yourself could make the warranty claim more difficult. The fact that reducing the power limit prevents the failure supports a thermal or power-related problem, but it does not prove that repasting is the correct fix.
Be cautious about treating the HWiNFO or GPU-Z hotspot value as definitive. Some newer GPUs do not expose all temperature sensors consistently through third-party monitoring tools, and the reported value may come from a vendor-specific or partially supported sensor. It would be worth checking the exact sensor names, updating the monitoring software, and comparing with the manufacturer’s diagnostic tools if available. Even so, the repeatable loss of output under full load is a legitimate hardware fault regardless of whether that 107°C reading is accurate.
I was not aware that hotspot reporting could be limited on the 5090. I used the default GPU Hot Spot Temperature readings in HWiNFO and GPU-Z, so I will verify what sensor they are actually reading before drawing conclusions.
The seller is a general second-hand store and does not work on PC hardware, so they cannot repaste it. My realistic choices seem to be requesting a refund, paying a qualified third party, or taking the warranty risk and doing it myself.