For some 2500h of GPU usage, its temperatures remained reasonable. Recently, however, they and RPMs increased dramatically. dGPU temperatures above 115°C are said to be dangerous for the GPU life. The hot spot temperatures of my GPU rose to 105°C and, during subsequent weeks, beyond. RPMs used to be below 1250 (hardly noticed) but rose to 1650, 1750, 1900 and eventually even 2050 (could be heard among the other PC fans). So I decided that a repasting was needed.

Turning the PC off, disconnecting the PC and GPU power cables and taking out the GPU was, of course, the easy part. Inspecting my GPU, I saw 17 relevant ordinary screws and 2 cables to the fans and lighting, 4 immaterial screws to ports and several specialised screws, which would only become relevant for disassembling the main board. I took off 10 screws on the back, 3 on the side near the ports, finally 4 screws of the backward die holder cross and the 2 cables. For every cable, one needs to press a switch while pulling the plug. For the 2 plugs on the GPU, gently apply flat pliers. I could loosen the shroud with the fans fairly easily.

The cooling body, however, would not separate from the GPU die yet. Unnecessarily, I feared having to buy specialised screw drivers. It turned out (and I had been warned in Youtube videos) that the thermal pads on the VRAM modules and VRMs behave a bit like adhesive. I put a flat screw driver between cooling body and main board near the die, turned it, cautiously applied pressure and repeated this process at four corners. Eventually, I separated the cooling body from the die with all the thermal pads remaining intact. Otherwise, one needs to buy and apply new ones of different thicknesses.

Most of the prior thermal paste was around the die, only fragments on the die with quite a few gaps and parts on the cooling plate. No wonder the temperatures were too high. I expected having to use Isopropyl / Isopropanol but soft, steady tissues applied slowly and generously sufficed for cleaning.

A hair drier warmed up the die and cooling plate. I used Thermal Grizzly Duronaut because this thermal paste got a good review. I put a 3x3 blob on the die and spread it cautiously, evenly, slowly and preferably without air bubbles with a spatula so that the entire die was covered but reasonably thinly. Afterwards, after 3 hours of work (my first repasting!), I reassembled everything.

The impact of the repasting the GPU die after ~3200h of GPU use is as follows. Maximum GPU Values when Running Katago for so far 18:40 after the repasting are:



This looks like a success. Even better, the RPMs are lower than I recall having ever seen before. Of course, I need to watch values during the coming weeks.

Watch the hot spot temperatures of your GPU. If they are unavailable, changes to RPMs are also a good indicator. Therefore, know early RPMs shortly after purchase and compare them to current RPMs. Much higher values mean that the GPU tries to protect itself but understand them as a reminder to repaste soon. Fast chips need regular care, as fans need regular dust cleaning.

Those temperature drops are insane, especially the hot spot. Thanks for sharing all the specifics.

There is a bit of increased variation up to the maximal values of the first week after repasting:



Mostly, the new temperatures are ~64.5°C and ~79.7°C. Note that, according to HWinfo64, my GPU firmware has always had the target temperature limit 84°C, which now it can fulfil again. Having looked up my first data shortly after purchase almost two years ago, it was 1078 RPM so I am back there again.