I have a five-year-old ASUS TUF Gaming OC RTX 3080 Ti that has never been opened or repasted. It is paired with a Ryzen 7 9800X3D and a new 850W power supply. When games run at high frame rates, the GPU driver completely crashes, the fans jump to 100%, and I have to hard-reboot the system. The core temperature is only around 50–57°C when this happens.
I have ruled out the PSU, since the 12V rails remain around 12.13V and 12.03V until the driver failure. VRAM junction temperature peaks at only 78–88°C, and the core never exceeds 59°C.
However, HWiNFO recorded one crash at 52.6°C core temperature and 98.1°C hotspot temperature—a 46°C delta. Since the fan curve follows the core temperature, the fans were only running at about 1,300 RPM while the hotspot was near 100°C.
As a workaround, I am running 0.9V at 1800MHz, an 80% power limit, and a manual 80% fan curve. That lowered the hotspot from 98°C to around 80°C and reduced the delta to 22–24°C. Power consumption dropped from about 336W to 263W, and the card is stable, but I lose performance and the tuning keeps disappearing after reboot. If I remove the undervolt while already in a game, the system crashes within roughly two seconds.
Is a 46°C core-to-hotspot difference most likely dried-out or uneven thermal paste, or could another problem cause it? Would repasting or using a material such as PTM7950 likely fix the card, or could the GPU already be degraded? I have never opened a graphics card before. If replacement makes more sense, what would pair well with a 9800X3D for 1440p gaming, and is it worth buying a new card now?
3 Answers
A 46°C delta is not absolute proof of bad paste, but after five years of thermal cycling it strongly points to poor contact or paste pump-out. The card is probably not permanently degraded. Cleaning the die and cooler and applying PTM7950 would be a good first repair; it generally handles repeated heating better than ordinary paste and could reduce the hotspot by 20–30°C. Be careful when opening the card and follow a teardown guide for the exact model.
A five-year-old RTX 30-series card is a good candidate for fresh thermal material, especially with such a large hotspot difference. I would service the cooler before assuming the silicon has failed. Replace the core interface material first, then check the hotspot delta and stability under the same workload before spending money on a new GPU.
I would try the repaste before replacing the card. The undervolt reducing the hotspot and making the crashes disappear is consistent with a contact or cooling problem, though it is not definitive. A bad mounting surface, shifted cooler, or uneven pressure could produce the same symptom, so inspect those during the repair as well. If you decide to upgrade anyway, a Radeon 9070 XT would be a reasonable 1440p pairing for a 9800X3D, but there is no urgent need to replace the 3080 Ti if a repair restores normal temperatures.
So I should replace the material on the GPU core itself, but leave the thermal pads and VRAM alone unless they are damaged or no longer making contact?