Thermal Performance

I will briefly describe my own prep work here. Upon seeing the two pieces separate, I was first struck by how thick that silicone glue is. I measured the green substrate PCB alone at 0.77 mm thick and the IHS and PCB together at 4.27 mm thick before de-lidding. The first thing I did was to use a thin scraping tool I had from work to gently scrape off the silicone, and I then used a cuticle to collect some of the stock Intel TIM for testing before using isopropanol on cotton swaps to clean off any remaining TIM. This was done for both the die and IHS, which also showed there was decent contact made between the two. If you do not have a scraping tool, an old credit or business card will do the trick as well. If you go with the Rockit tool, the provided wooden cuticle helps too.


Continue to scrape and wipe until you have little left on both the IHS and PCB; do not worry if you are unable to remove the last bits off. More important is making sure it is uniform and that there are no fingerprints or finger oils on the PCB or die. If there are any, be sure to wipe them off before re-lidding. Once done, mask off the PCB around the die using non-residue tape and apply your TIM of choice on the die. I used Thermal Grizzly Conductonaut for the best-possible results here, and since it is an electrically conductive paste, you might want to insulate any exposed traces or components, including those four traces to the bottom left of the die in the last picture above. Any electrically insulating material, preferably as thin as possible, will work fine here - a thin layer of liquid electrical tape, for example.

Then it was off to re-lidding. I used the Rockit Cool approach and a drop of Loctite brand super glue on each corner only as opposed to more silicone glue between the two. After all, my aim was to not only use a higher conducting TIM but to also reduce the thickness of the heat transfer layers. For the sake of testing this second variable, I did a further de-lid and applied silicone glue as well. I let the IHS be under pressure and in contact with the CPU PCB for five hours to ensure the job was done.

In order to test the effect of de-lidding, I used the same testing procedure as described here. Of course this being a different CPU entirely, I went with the i7 7700K at 4.9 GHz and 1.32 Vcore instead. As such, the results from this test can not be directly compared to my other CPU thermal performance results. Instead of average CPU core temperatures, I also recorded the CPU package temperature and monitored ambient temperatures to get a delta T (CPU package - ambient). An EK Supremacy Evo CPU block with a Laing D5 Vario at Setting 3, a Black Ice Nemesis 360 GTX, and three EK Vardar F4-120ER fans at 1500 RPM were chosen for cooling the CPU. The choice to go with a custom loop was to get the best-possible results without going way overboard, and I feel the subset of watercoolers and de-lidders is fairly large anyway.

Finally, a crude house-built thin film conductivity measurement device at my university was used to get an approximate thermal conductivity value for the stock Intel TIM scraped and collected from the IHS and die. Not the best way to do it and certainly not how I would have preferred to do it, it will suffice as an indication of where things lie. Here are the results below:



These are absolutely massive decreases; 20.7 °C from the TIM swap and fresh batch of silicone glue and 24.4 °C from the TIM swap and super glue. By going with a drop of super glue at each corner instead of a layer of silicone glue in between, I reduced the distance between the CPU die and IHS and, thus, also reduced the heat-transfer distance, which aided heat transfer away from the die. So there are two variables at play here which both affect the internal heat transfer from the CPU die to the IHS, and you want to work on both for the best-possible results. Take these numbers as an indicator rather than absolutes as I am sure I could have achieved better results with the silicone glue run, and there is also no guarantee my liquid metal spread was consistent both times. Similarly, the measured thermal conductivity value for the Intel stock TIM was 3.1 W/mK. Note that this is more of an indication of where things are at, but it is not as bad as many would have you believe. Most stock TIM provided by manufacturers of coolers and waterblocks alike are in the 2-4 W/mK range as is. But Thermal Grizzly Conductonaut has an advertised thermal conductivity of 73 W/mK, and this is a big, big part of that 20+ °C drop in temperature.

Finally, I will also mention that I was not comfortable with the CPU package temperature before de-lidding for when I aimed to hit the 5 GHz mark. But after de-lidding, I am getting lower temperatures at 5 GHz under load than I was with no overlock applied at all.