Today the reviews of Intel's Core i9 9900K have gone live, including our own, and something has been noticed by some reviewers, and that is these chips can get pretty hot when overclocking. After all of the criticisms of Intel for using a paste for the thermal interface material (TIM) between the die and integrated heat spreader (IHS) for previous generations of CPUs, the companies move to a solder TIM seemed like good news, so why is this? Extreme overclocker der8auer decided to investigate and shared the results on his YouTube channel, where he not only delids the CPU but even removes part of the silicon carbide layer though lapping to improve performance.
Near the beginning of the video, der8auer states that when he first test the 9900K he was impressed by how hot it got, reaching 90 to 95 ºC on a 360 mm AIO cooler. He was expecting it to much lower and when looking to power consumption, it still seemed rather high, so after reaching out to others he decided to delid the CPU. As the thickness of materials impacts their thermal conductivity, he decided to re-solder the CPU, but with a thinner layer to hopefully improve temperatures. Unfortunately this test failed as it appears the solder did not connect the die and IHS well enough, so it appears the thickness Intel is using is what is needed. Unsatisfied though, der8auer continued by trying liquid metal, finding it improved temperature some (from 93 ºC to 84 ºC at 4.8 GHz with 1.25 V), and then he decided to go even more extreme. What is more extreme than delidding a soldered CPU? Lapping the CPU die to remove some of the silicon carbide layer meant to protect the circuitry beneath. It should be noted he was using Prime95 for his testing, which will stress a CPU very well, but is not always indicative or realistic loads.
When physically comparing the 9900K to the 8700K, he noticed both the PCB and the die is thicker, at 0.87 mm compared to 0.42 mm. Remembering the thicker a material is, the worse its thermal conductivity is, he decided to take an i5 9600K, delid it and then grind down the die first by 0.15 mm and then by another 0.05 mm, for a total reduction in height of 0.20 mm. Without these modifications, the 9600K reached 96.5 ºC when running at 5.0 GHz using 1.35 V, while delidding and using liquid metal dropped it to 88.5 ºC. With the die lapped to remove 0.15 mm the temperature was 84.66 ºC, and with 0.20 mm removed it was 83 ºC.
With these results, der8auer is left wondering why the die is so thick compared to previous Intel CPUs. He wonders if perhaps the thickness is needed for soldering to be successful or if it might be needed because the die itself is larger than that of the 8700K. If it is because of the use of solder, he then questions if the criticisms on Intel to use it instead of paste could have worked against the user in the end. (By the way, as I recall it was not the use of the paste itself, but how much Intel used.)
Source: der8auer YouTube (embed above)
Back to original news post