Modifying Gigabyte's Windforce 5X Cooler
The Vapor Chamber: Just A Tad Too Small
Even though the Radeon HD 7970 Super Overclock's vapor chamber is huge, it’s not quite large enough. It’s easy to see that it was originally designed for Gigabyte's GeForce GTX 680, and would have needed about 1 cm more to properly fit the 7970.
Just look at the large yellow rectangle in the picture below. You can see that the two memory packages at the bottom aren’t completely covered. But because the vapor chamber doesn't touch the RAM at all (even the ICs that are covered), Gigabyte uses thermal pads on them. This turns out to be counterproductive, since the pads isolate the memory from circulating air.
Not surprisingly, we saw graphics glitches when we overclocked the card’s RAM. They went away after we scraped off the thermal pads using a box cutter. Even if you don't go to that extreme, you should be alright, so long as you keep the memory running below 1500 MHz.
The heat sink does a better job of cooling the card's voltage transformers and coils, which have their own aluminum plate dedicated to drawing thermal energy away.
Thermal Paste: Room for Improvement
The Radeon HD 7970 Super Overclock's thermal paste proved to be thick in both consistency and application. After an initial round of (disappointing) temperature benchmarks, and after talking to Gigabyte, we decided to replace its compound with our own and repeat the benchmarks.
We settled on Gelid Supreme as our thermal paste of choice due to its easy application and short time to reach maximum transfer performance. We also found that four of the six screws holding the cooler’s frame either hadn’t been screwed in tightly or came loose in shipping.
It was important to us that the thermal paste stayed elastic, even after being exposed to high temperatures, so that it could handle any possible board flexing. Prolimatech’s thermal compounds have worked well for us in the past, but they need a longer burn-in time to reach their peak performance. We put the card through a four-hour stress test before running our benchmarks.
The following tests weren't conducted with Gigabyte's thermal paste, but with our substitution. We believe that the card would have shown similar results to our own had it not encountered transport problems and suffered incorrect application of its original thermal compound.