I bought a second-hand Gigabyte AORUS RTX 5090 about a month ago. When I play demanding games such as Doom: The Dark Ages, the GPU eventually stops outputting video. The rest of the computer keeps running—the game audio continues and my peripherals remain responsive—but the display does not return until I perform a hard reset.
OCCT found no VRAM errors, and the same behavior occurs on both CachyOS and a fresh Windows 11 installation, so it does not appear to be driver-specific. Monitoring with HWiNFO and GPU-Z showed the core temperature around 66°C while the reported hotspot reached approximately 107.4°C, a difference of more than 40°C. The signal usually cuts out shortly after the hotspot reaches that level, although the fans continue spinning.
Reducing the power limit to 70% in MSI Afterburner prevents the crash and greatly reduces the reported temperature difference. This makes me suspect an uneven cooler contact or thermal-paste problem, but I am not experienced with GPU repair. I bought the card from a general second-hand store in Japan, and it is still covered by their warranty. They do not specialize in PC hardware and cannot repaste it themselves.
Should I return the card and request a refund or replacement, find a third party to inspect or repaste it, or consider doing the job myself? How risky is repasting a high-end GPU, and could the reported hotspot reading be inaccurate on this generation of card?
3 Answers
Repasting a graphics card is manageable for someone experienced, but it is not a harmless beginner repair—especially on an expensive 5090. The thermal pads and cooler mounting pressure need to be restored correctly, and a mistake can make the temperatures worse or damage the card. If the original seller cannot handle it, look for a reputable GPU repair technician, but returning it is safer while the warranty is available.
Before treating 107°C as definitive proof of a bad paste job, verify what the monitoring software is actually reporting. Some newer GPUs do not expose every temperature sensor through ordinary third-party tools, and software may label an available sensor as hotspot. Still, the fact that lowering the power limit prevents the crash is useful evidence of a power or thermal issue. Check the power connectors, PSU capacity, and physical card sag as well, then use the warranty rather than modifying the card yourself.
I was not aware that the 5090 might not expose the same hotspot data as older cards. HWiNFO and GPU-Z both show a GPU Hot Spot Temperature value, so I would like to confirm exactly which sensor that value represents before deciding whether the cooler needs to come off.
Since the card is still under warranty, returning it or requesting a replacement should be your first choice. Repasting requires removing the cooler, and you could damage the board, misplace thermal pads, or give the shop a reason to reject the warranty claim. A large core-to-hotspot difference can indicate poor cooler contact, but the card should be inspected before you take that risk.
The seller is a general second-hand store rather than a PC hardware shop, so they cannot perform the repaste. My practical options seem to be returning it, finding an independent technician, or accepting the warranty risk and doing it myself.