We tested 90 different thermal pastes to help you find the best one to help cool your CPU.
Best Thermal Paste for CPUs 2021: 90 Pastes Tested and Ranked : Read more
I've been using Nocta's paste pretty exclusively for the last decades, because it came with their coolers, which I've also used pretty much exclusively for about as long. The performance of both has typically been rather well, was long as the CPU's Wattage and the cooler variant were matched.
I like Mini-ITX as a form factor to build µ-servers, they are not much bigger than NUCs yet offer a lot more flexibility for building your own and recycling components.
I also like the 28-45 official TDP range for SoCs, because those µ-servers typically run 24x7 and energy/heat is an issue and because modern laptop SoCs can deliver quite a punch without going to the extremes of 250 Watt+ peak power draw.
But only the more enterprising Chinese manufacturers seem to deliver that combination of small and powerful, often after Intel starts discounting last generation chips.
Typical case in point are Erying boards, in my case a Mini-ITX G660 with a soldered i7-12700H, two DIMM slots and plenty of expandability via M.2 slots.
They come with a custom heat-spreader to cover the naked mobile die and make it height-compatible with a typical desktop cooler expecting a lidded chip. Since it's officially a 45 Watt TDP SoC and the chassis is the typical minimal Mini-ITX case I went with a Noctua NH-L9i-17xx fan, which is supposed to cover up to 65 Watts.
I've used that and its slighty bigger brother NH-L9x65 in a lot of builds for systems in the 35 or 65 Watts range respectively and found it rather capable of handling peaks in a very relaxed manner, using the relatively high mass of the cooler to store heat, which was then dissipated over time by the relatively low air-flow fan (it's small and thin after all).
The system was rather disappointing and somewhat puzzling, it never seemed to get hot, rarely reaching 70°C yet it signalled thermal limits nearly all the time, never going much beyoned 25 Watts at sustained loads, according to HWinfo.
I replaced the orignal paste between the naked SoC surface and the Eryin heat-spreader with Noctua's paste (also used between the heat-spreader and the Noctual cooler), but that didn't really do a lot.
Since there was such a lot of bad press about Intel using paste between their naked chips and the heat-spreader I thought it a good opportunity to test liquid metal instead of paste, even if the 4°C range cited here didn't exactly hint at a giant difference.
The other motive was that the SoC doesn't have any capacitors or contacts near the die, actually features a bit of a basin around the die carrier and the system is operated "sunny side up" (actually down), meaning there would be no risk of liquid metal seeping onto the motherboard creating shorts.
I've never seen such a night-and-day difference!
I'll freely admit that I typically only pasted on top of heat-spreaders, but I can't have done a completley incompetent job either, because the Noctua paste decreased the original temps at least a little bit.
But with the liquid metal, the SoC was unrecognisable! Where the fan would spin to its max 2600rpm at only 25 Watts before, it now stayed at a rather less noticable 2200rpm while the SoC kept going on its default PL1 limit of 95 Watts--on a cooler that is rated for 65 Watts! It just never slowed down and temperatures stayed around 75°C, too.
Now those temperatures had actually been indicated at slightly less before via HWinfo, not even reaching 70°C. But I am pretty sure now, that those were "false readings" in a way: the SoC would overhead so rapidly on paste, that it had to throttle brutally, resulting in the 25 Watt consumption but also rather "mobile" clocks.
Now with Conductonaut liquid metal the SoC could actually go to its full potential and never throttled.
I also applied it between custom heat-spreader and the Noctua cooler, which might have had less of a dramatic effect, but since there was plenty left over and the spill risk due to the upright position remained minimal, it seemed the best thing to do.
PL2 turned out the next challenge, because the Pico-PSU only supplies 120 Watt maximum and any Alderlake will take a huge draw from the power bottle, unless it's constrained. And there are some YouTube videos which show that the power delivery circuitry on the board is simply not matched to the 168 Watts the BIOS allows for PL2.
But constraining the power inputs to something that fits the SoC's official range and the board design (I settled for 65 Watts sustained) is a snap and now has the fan remain below 2000 rpm, completely unnoticeable underneath my table as a side benefit. Peak performance is hardly impacted and for sustained workstation loads I have workstations.
My major conclusion is that
in any naked die scenario, the benefits of liquid cooling vs. paste are way bigger than between heat-spreader and cooler.