Test System

• All testing is conducted with the ASRock Radeon RX 7900 XTX Taichi at its maximum power limit setting (+15%)
• The heat load is a constant game-like TPU in-house load that's crafted in a way that lets us control the exact wattage level that is applied to the card
• Power is measured externally using test equipment, with 40 samples per second, the reported number is "full card"
• Temperature is measured with GPU-Z using the graphics card's own temperature sensors, we've recorded both "GPU temperature" and "Hotspot temperature"

Performance

We're starting our load testing with just 100 W and will ramp it up to the card's maximum of around 475 W. The power draw of the card, which equals the heat load on the cooler, is plotted on the x-axis, the temperature is on the y-axis.

For each TIM we're doing two full runs, between runs the paste is fully cleaned and a new paste from the syringe is applied. For the Thermal Grizzly KryoSheet I used a brand-new pad each time.


When looking at the "GPU temperature" sensor, there is only a very small difference between all materials tested. While the differences can be in the order of 1-3°C, such numbers are pretty close to the margin of error, which I'd estimate at 1-2°C. I've applied thermal paste hundreds if not thousands of times in my life, which reduces the application error, but it can't be eliminated completely.

For the KryoSheet, things are looking better than expected. I assumed that it would perform slightly worse than the pastes, but that's not the case. Its thermal performance is really good and virtually identical to the other pastes in our test group.

Please note that chart's Y-axis does not start at 0°C. This is intentional, to give us a little bit more detail, otherwise the separation would be even smaller and the pastes would appear to be even closer to each other in performance (yes they are, I'm not trying to misrepresent anything).


On AMD Radeon GPUs, there is not only the classic "GPU temperature" sensor, but also a sensor called "Hotspot," which represents the highest temperature reading from a large number of temperature sensors located throughout the GPU chip.

Here we got some more-interesting results: the first run with the KryoSheet (dark blue line) shows surprisingly high hotspot temperatures at lower heat load ranges, but for 300 W and above, the temperatures are fine, and at maximum heat load, the KryoSheet actually achieves the best temperatures of all our TIMs, by a small margin though.

I have the following explanation. In the first run, the KryoSheet didn't make optimum contact with all areas of the GPU die. The area that wasn't cooled so well ran at higher temperature than with the other pastes, which caused the hotspot to move into this specific section, because the rest of the chip was cooled to lower temperatures. Remember, the hotspot is not one specific sensor, but the hottest of all the sensors of the GPU, which means it can "move." As we increase the heat load, other parts of the chip get warmer, like they usually do, and now the hotspot jumps to report the temperature of this, now hottest area of the chip, which makes very good contact.

It's possible that my first cut of the KryoSheet was a little bit too small, or it moved a bit during installation for the first run, so I recommend you don't cut it to the exact size of the die, but leave a little bit of overhang—the capacitors are not that close.

Compared to the "GPU Temperature" readings, the "Hotspot Temperature" results show a little bit bigger differences between the various pastes at maximum heat load. It's still not a night-and-day difference, but I think 5°C is something that you can start appreciating, especially on a higher powered card like the RX 7900 XTX.