We live in a unique time where the process and performance race has reached a new boiling point. 10 cores, hundreds of megabytes of L3 cache, 5 GHz and other decisive factors make fans of both camps face off again and again on specialized forums, proving who has the coolest overclocking or benchmark results. A simple user who wants to do something like this, getting into this whirlpool, just starts to get lost from the amount of advice and opinions. Of course, every idea and every opinion has an undeniable right to exist, but today it will not be about that. I want to discuss with you the origins of cooling problems and a modern solution using the example of a line of water blocks from EKWB. In particular, we will discuss how to cool hot, but economical (in 2019 it did happen) third-generation Ryzen processors based on the Zen 2 microarchitecture.
As many of you know, a processor in a section resembles a sandwich: it consists of a textolite, with which the crystal is connected to the socket, the crystal itself and, of course, a thermal interface and a heat-distributing cover that take all the heat shock. The second important information that you may not have taken into account is heat dissipation relative to the processor die area. Of course, with a constant TDP, a smaller area will create more problems for absolutely any cooling system (we are not talking about «freons» and liquid nitrogen). As for the numbers, we have the following data:
- AMD Ryzen 7 2700X with a die area of 213 mm² and a TDP of 145 W has a heat flux of 0.68 W/mm²;
- Intel Core i9-9900K with an area of 174 mm² and a TDP of 200 W — 1.15 W / mm²;
- AMD Ryzen 7 3800X with a CCD area of 74 mm² and TDP 90 W (excluding IOD) — 1.21 W / mm²;
- Intel Core i9-10900K with an area of 198 mm² and a TDP of 250 W — 1.26 W / mm².
We see that the latest generation of Intel and AMD processors have significantly increased the heat flux generated relative to the area, and something needs to be done about this, because a further transition to a thinner process technology will only aggravate the situation, besides, silicon is not a good heat conductor. For example, the thermal conductivity of silicon is 149 W / (mK), and copper is 401 W / (mK).
Intel has found a way out of this situation, and this is not only at the cost of solder. At the end of spring, the Comet Lake processor line was introduced, an interesting feature of which is the “grinded” top layer of the silicon substrate and, along the way, an increased thickness of the heat-distributing cover.
In numbers, it looks more interesting, Coffee Lake has a silicon substrate thickness of about 800 microns, and for Comet Lake-S processors it has decreased to 500 microns. The difference is almost 38%, which certainly worked and paid off — 5.4 GHz for all 10 cores in home use without any chillers. For ordinary users, this innovation made it possible to see lower temperatures, which is good news.
As for further potential, additional “polishing” is still possible.
Since the BEOL layer still has an impressive thickness and the only limiting nuance is the technology of this very grinding, because our favorite transistors are on the border of FEOL and BEOL.
What about AMD? At the moment, we have only one 7nm generation, which has already managed to establish itself as a very energy efficient solution, but at the same time hot for rather primitive cooling systems: whether it is maintenance-free water or a direct contact cooler.
I propose to consider this in more detail, in particular, the water block of an unattended dropsy. It consists of a heat-distributing copper plate (sometimes nickel-plated), in which there are slots (an internal radiator), and, in fact, a pump. If everything is clear with the pomp, then there is often no water block with the camera, since users rarely have the opportunity, without violating the guarantee, to open and examine the insides on their own. And if such an incident happens, the user sees that in his favorite non-separable dropsy, the chamber with microchannels does not cover even half of the area of the processor cover. In the illustration, it often looks like this:
At first glance, the active chamber of the water block covers both IOD and CCD, but physics is often overlooked, heat does not spread vertically upwards, but evenly in all directions relative to the emitter (for now we do not take into account that other surrounding materials that have their own mass and thermal conductivity). In this regard, we need to remove heat from the entire processor cover (IHS), since it plays the role of a heat spreader. The ideal option would be a significantly enlarged water block chamber with microchannels:
Fortunately, such a product appeared in the EKWB water block line, and today we have the opportunity to compare theory with practice, whether this gives any advantage and, if so, what.
The last and probably the most important characteristic that a water block should have is the efficient use of all microchannels and water flow, respectively. Let’s consider an example of using a typical cheap water block on processors from both camps.
The first thing that catches your eye is the orientation of the CPU chip(s) and, accordingly, the transverse arrangement of the microchannels relative to the chip. If the microchannels are longitudinal — 1/2 of the flow will be used suboptimally, in particular, the situation will worsen on two-chip Ryzen, where 1/2 microchannels (and 1/2 of the flow) account for two serious heat sources. This leads to the conclusion that the water block created for the Intel processor (before the release of Zen 2 there simply were no other options), due to a number of physical laws, is not able to work with maximum efficiency, although, of course, it will work, since heat spreads in all directions, depending on thermal conductivity of materials surrounding the heat source and solder under the IHS processor, which is also a kind of heat spreader plate. An exception may be a water block with a massive copper heat exchanger, which in turn works as a “sponge” of heat, or a water block with a brass (or copper) chamber and with an offset inlet flow, which will point to the location of the chiplets (CCD). A prime example of this design is our new Magnitude.
In the case of Intel, we don’t see much magic, but with Ryzen, we can see that a concentrated stream of cold water hits the intersection between two CCDs and creates additional pressure in the microchannels (marked with purple arrows).
It sounds good, but I still found one nuance that to some extent also affects the result, because due to the small mass of the IHS processor, in the case of Ryzen, it is not evenly warmed up, the flows that go through the IOD (allocates up to 20 W), have low efficiency and simply fall into the outlet. Perhaps this nuance was left for the future generation of water blocks, because the further development of water blocks has almost reached the point where there is simply nothing to improve and a completely different way of cooling processors needs to be developed. And now let’s get to know the hero of the occasion closer.