Date: Tuesday , February 07, 2017
For those of you just joining us, here is a bit of history leading up to this quick article. We started off with delidding and relidding the Intel Core i7-7700K and then moved to the Intel Core i5-7600K.
And finally the article that spawned this follow-up.
In that article and supporting video we were able to get our 7600K running without its Integrated Heat Spreader installed, aka "naked."
To be succinct, I would not suggest going naked with the die at this time with Kaby Lake or Skylake processors. The best temperature delta I was able to get out of the naked die was 3 degrees C. Not exactly a huge benefit with the risk involved with going this route.
While I can highly suggest a delid and relid, which has its own risks, you are diving into a whole other pool of possible nastiness with running the die naked. First and foremost you are going to have to put your motherboard's socket under the knife, literally. As shown in the video above, the corners of the LGA 1151 have a Z height that is simply too tall to allow our water block to sit down far enough to properly hold the CPU into the socket with the hold-down clip removed. Of course when you start using a razor knife, or any other type of tool to modify the socket, you of course open yourself up to destroying your motherboard beyond repair. One little slip of the tool could very well cut a trace on the PCB, not to mention all the other components that are there to destroy.
Beyond actually damaging the motherboard, running the die naked puts us in a position where the CPU is "floating" more in the socket than it would be normally with the much wider and taller IHS mating surface. While we did not let any of the magic smoke out, this time, I would not doubt that it could be done depending on how "crooked" your naked die CPU is sitting in the socket. Permanently destroying your expensive CPU is likely not worth the small temperature decrease we saw.
In the past, naked die cooling was the only way to do it. Along with that we saw a lot of folks, ourselves included, that chipped processor die corners off, effectively killing our CPUs. This was however back when our CPUs had pins that inserted into the socket then locked down. Once those chips were locked into the socket, those basically no longer moved, and when tightening down a metal cooling surface to that naked die, if the mating was not "perfect," the force exerted would literally break the die.
That situation has fully changed with Intel's Land Grid Array (LGA), where the pins have been removed from the processor, and have now been transferred to the socket down on the motherboard.
As you can see above, there are not "pins" down in the LGA socket. These pins are actually what are better described as little metal fingers bent over a bit that form what is a spring. This is why I refer to the processor as "floating." While our cooling mating surface is pressing down onto the CPU die, the die can actually move. It is not static in position like pinned CPUs, and therefore tremendously lowers the risk of breaking the die with the cooler.
I was truly expecting to see better temperature results than we did with the naked die. I am Jack's broken heart. The problem as I see it is one that pertains to shimming the naked processor correctly. I felt as though I was never able to get the die properly mated to our cooling system water block. Since we do have two axis of control diagonally across the face of the die with our water block, I did spend a couple of hours trying to position the water block where it mated better to our die facing. To do this, I literally manipulated the screw/spring water block hold-down adjustments while the processor was running under full load and watched my per-core and package temperatures. I would NOT suggest doing this at home. However, with all my efforts I was never able to see more than a 3C temperature advantage with using the naked die cooling.
Kaby Lake (and Skylake) naked die cooling, while seemingly very attractive got us little benefit, especially when you take into account the risks associated with doing it. In order to get "perfect" surface mating we simply need a much more accurate shimming system than the seat-of-the-pants approach we used here. Even if we had that perfect shim, we are still going to have to carve on the LGA socket some, or have a very specific water block that will sit inside the edges of the socket walls, and I am not aware of one of those in production currently. My suggestion is go with a "delid and relid" using a liquid metal Thermal Interface Material, like CoolLaboratory Liquid Ultra, for the time being. We saw great load temperature results using that method with our 7700K, as well as our 7600K.
For those of you in the market, the Core i7-7700K is still available on eBay via Monoprice for $310 with no tax or shipping. I have two more on the way for more testing. The Core i5-7600K is going for $240 at Amazon with Prime Shipping. If you are primarily a gamer, the 7600K is the way to go.