How did we get here?
What do you do when your graphics card starts running warm and sounds like a jet engine? Set aside some time, trust yourself, grab your tools and it may help to watch a teardown video or two.
My adventure began when I started benchmarking the RX 9060 XT against some other cards that I have. The testing went well, but I noticed how loud and warmer temperatures for my 3060Ti. This card has been a workhorse for me , and I only recently took it out of my main PC. I didn’t notice anything as I was using it to game and stream, but it was very noticeable during the comparisons.
I also noticed that the NVidia card ran much warmer than the other two by several degrees. Benchmarks measure performance by stressing components, but this card had some abnormal behavior. It wasn’t hard to see I had a small issue.
The card I have, an EVGA RTX 3060Ti, is a two fan card with a slightly larger heatsink, but it was obvious the cooling on this card was working much harder than the others. In addition, the time to heat up was very quick and cooling off was more gradual. The card was in regular use since purchase, but very seldom was it stressed for long periods.
The next steps
At this point I was sure of a couple of things. The card wasn’t overheating, but was warmer than normal. It was time to make sure that the temps didn’t adversely affect the results I was getting in the comparisons. The only way to make sure was to tear it apart and reapply the thermal paste. I was just hoping the change wouldn’t skew the numbers I had already collected for the video I just released, here.
The teardown itself is straight forward. Several screws hold the backplate, but the ones near the GPU take special care. These four spring loaded screws are handled in an alternating pattern similar to changing a car tire. This ensures equal pressure on each side of the chip and less chance of damage. Standard thermal paste like MX4 works fine assuming you make sure coverage is full.
Testing
With the paste applied and the card reassembled it was time to re test all of the benchmarks. The strange thing is that in almost every benchmark the results were slightly lower than the first set of runs with a lower temp. So, what happened, it should have been better.
Actually, it was better, just not the final numbers.
Observing the tests, it was important to see not just the temperatures, which were slightly lower, but the noise levels and the ramp up and cool down. In the first set of tests, the card would heat up quickly and recovery was slow. In this set, temps rose gradually and cooled off very quickly. My first indication that the re-paste worked.
Other indicators were the sound of the fans during stress, and the sheer amount of heat put off. The first set put off a considerable amount of heat when opening the case, while the second test was closer to expected. The fans were much quieter as well. There was fan noise, but it not to catastrophic levels. This was a success.
Now What?
The most important thing is that my comparison to the other two video cards is still valid. The results from testing were still within margin of error, so the results of the first test were solid. The card is cleaner and with fresh thermal paste, it will have long life.
The next thing is that it really was something simple and well within my experience level to fix. There are several small thermal pads on the video memory chips, and of course the paste itself, but the hardest part of this was keeping up with the screws. Not only did the back plate have several, but the end bracket had three tiny ones and two for the back plate had very small nuts. (go ahead, insert joke here).
The only other worry was making sure I used a cross pattern to tighten the screws holding the PCB to the Heatsink. Those were spring loaded, so the chances of overtightening was slim, but there was still a chance. It pays to take your time and exercise patience here.
Overall, this was a great success. I wasn’t sure of that when I started seeing the benchmark numbers come back slightly lower, but the cards behavior was vastly different and I’m happy. This wasn’t a repair as much as maintenance, but it’s good to know it was successful.
The YouTube video for this blog can be found here.
Feel free to check out more blogs with other things I probably shouldn’t have done. here