Thresholds in computing: Part 9 – heat dissipation and Thin-ITX

(Part 9 in a series of posts on small-form-factor computing)

I wish I had the equipment to make the kind of heatmaps that Puget Systems does in their comparison of horizontal vs vertical cooling. But I don’t, so we’ll just have to make do with overlays again.

Heat sources in a passively cooled system
Heat sources in a passively cooled system

This is a very crude “heatmap” based on how hot various components feel to the touch—the redder, the hotter. It is, of course, not a direct representation of the amount of power consumed or heat produced by each component; it is obvious that the CPU is the main source of heat, but it feels cooler to the touch because of the huge CPU cooling block and the heat being ducted to the side heatsink fins by the heatpipes. These various components are:

  1. Wifi module
  2. PCH/Southbridge (“connectivity hub” for storage and peripherals)
  3. SSD
  4. Voltage regulators (converting supplied power to 1V for the CPU)
  5. Main cooling block (For CPU)
  6. Memory modules (SODIMM)
  7. Internal AC adapter for 19V output
    Of these above components, the ones that feel hottest to the touch are the Southbridge aka Peripheral Controller Hub aka PCH (#2), voltage regulator modules aka VRMs (#4), and the solid-state disk aka SSD (#3). In its Thin-ITX System Design Guide, Intel also lists “the CPU, voltage regulator components, SO-DIMM memory and the PCH” as the primary components for consideration when it comes to cooling.

As its name suggests, the PCH is the gatekeeper for data throughput to/from the peripheral devices (i.e. anything that is not a processor core, main memory interface, or graphics chip)—this means the network chip, audio chip, storage interfaces, USB, etc. Intel’s spec page for the Q87 chipset lists a max TDP of 4.1W, which is close to that of a mid-/high-end smartphone.

The VRMs are the voltage regulators that convert 12V supplied to the board to 1V required by the CPU. This conversion is inefficient and produces a lot of heat, and these components are a major cause of concern for overclockers. They are also found on graphics cards, any place where voltages need to be down-converted to processor operating voltages.

Lastly, the SSD. SSDs come in various performance ranges, and hence also a range of TDPs. It’s not easy to find expected temperatures for each SSD, but the higher-power-consumption ones also tend to dissipate a lot of heat.

Heat flow in a PC #

When considering the cooling system of CPUs, it is helpful to think of heat as being a flow of energy (not technically/scientifically correct, but the simpler mental model is easier to imagine). The various primary heat sources mentioned above dump all their heat into the air in the case, causing it to heat up. Of course, this can’t happen indefinitely, something has to take place to remove this heat.

The primary problem here is how to transport the heat away from those heat sources faster than it can be produced. If you have taken high school science, it should not be foreign to you that heat is a form of energy and can therefore be measured in joules. In a fixed amount of time—it is helpful to think of 1-second intervals—all these parts consume a certain amount of energy, use some of it to do useful work (tunnelling electrons through barriers, moving hard drive/optical drive parts, transmitting wireless signals, etc), and dissipates the rest of this energy into the case air as heat—all energy taken in must go somewhere.

So in any given one-second interval, the PC dumps a certain amount of heat energy into its air (inside the case). It dumps heat more slowly when it is idle, and more quickly when it is under load. If you are building a PC, you need to know the rate at which each component will dump heat into its surroundings when it is under maximum load—this is called the component’s Thermal Design Power, or TDP (non-geeks out there, sorry for taking so long to explain this). Since we are measuring the time rate at which energy (in joules) are being dumped into the surroundings (inside the case), TDP has units of watts (W).

Unfortunately, there is no industry-standardised way to measure TDP. So an Intel processor rated 84W TDP by Intel may not be dissipating heat at the same rate as an AMD processor with the same TDP rating given by AMD. Nonetheless, in is in the nature of geeks that when they see two numbers that look similar they can’t help putting them side by side anyway.

Temperature and heat dissipation limits #

So how hot can PC components get? Theoretically, if heat is produced in a component faster than it can escape, there is no real upper limit to its temperature. The temperature simply keeps rising until one or more parts in the component fail, or until we hit the melting point of something and it begins to melt… this is a known problem with some earlier CPUs, but these days modern CPUs have thermal control built in and will simply throttle until temperatures are back in a safe range.

The operating temperatures for consumer processors is typically in the 70–100°C range (look up the spec sheet for your favourite processor to be sure). Desktop processors tend to have lower maximum operating temperatures than mobile processors (e.g. laptops). Unassisted, processors can dissipate heat from their own packaging at a rate of ~4W (incidentally, the maximum power consumption for most phones seems to be capped at 4W as well …). If aided by a large heat spreader or direct contact with a larger metal surface (e.g. in many mobile tablets), this can go up to 8W.

Any higher than this, and we need larger and larger heat sinks. Remember that the goal is to get hot air out of the case; dumping the heat from the processor to the case air will not suffice. So we need something that can move the hot air out of the case. Enter the humble fan.

Active air cooling #

Case fans in their natural habitat []
Case fans in their natural habitat []

An engineer will tell you that a fan is different from a pump; a fan moves gaseous-state matter while a pump moves liquid-state matter, and so on, but for the sake of this post let’s just treat pumps as devices that help move things along with the aid of electrical power.

The role of the fan is simple: to move air from one place to another. In its typical roles, it either forces air through the CPU heatsink to remove heat from the CPU more quickly, or it moves hot air from inside to outside the case. Airflow engineering experiments show that it is generally more effective to suck air out of the case with fans, then allow fresh air to enter naturally, rather than trying to blow cool air into the case and let the hot air be forced out naturally.

When this heat flow is assisted by pumps, we call it active cooling.

This is the default heat flow direction for typical computer builds. It is simpler to design, and the industry has optimised itself to produce PCs with this kind of airflow. What other alternatives are there though?

Water cooling #

Again, let’s keep our primary purpose in mind: to move heat from the PC components (namely the CPU) to the outside air. Heat by itself is a form of energy (vibrational/rotational energy of molecules), and can’t move on its own; it needs a medium of transfer. We have a name for the category of heat transfer mediums: coolants. Air is one; not a very effective one, but the most widely available, and the cheapest.

The ideal coolant should have high heat capacity—a small amount of it should be able to hold a large amount of heat, so we don’t need to pump a lot of coolant to remove heat. It should be easy to pump, which means it should flow easily, which means it should not be too heavy/dense or too viscous. Air flows easily, but has notoriously bad heat capacity and heat transfer capabilities (why do you think we use air layers to insulate our homes and our clothing?)

Enter water. Water is heavy/dense, but has some of the highest heat capacity of any liquid and also flows relatively easily (if pipe diameters are not too small, otherwise we run into severe surface tension effects). How can we use water to do this heat transfer instead?

Corsair closed-loop water cooler + radiator
Corsair closed-loop water cooler + radiator

The simplest way to do water-cooling today is with a closed-loop cooler (“closed-loop” simply means we don’t have an open coolant reservoir that we can use to top up the coolant. Again this is not a technically accurate explanation).  Liquid is pumped by an integrated pump (round object, top right, installed above CPU) in a loop, from the CPU to the radiator (top left), and back again. Heat is transferred from the CPU to the coolant though a metal conducting plate, and subsequently from the coolant to the radiator, where it is cooled by air being forced through the radiator fins. We skip the whole heat-being-transferred-to-case-air step here, pumping the heat directly to a radiator mounted near the outer surface of the PC.

We’ve been using the word “coolant” so far because we can use any liquid in that system, really. But usually we use water plus some additives, because water is a decent enough coolant on its own (the additives are for anti-corrosive properties and such).

One thing that active air cooling and water cooling solutions share in common is bulk; they take up lots of space! Most case builders, even the ITX builders, will disagree—but I’m coming from a perspective of trying to minimise empty space in the case, so humour me a bit. Closed-loop coolers have unwieldy pipes that can’t be bent too sharply, and air-cooling heatsinks need to be big to be of any real use. What other alternatives do we have?

Passive cooling #

Let’s talk about heatpipes. It is not an exaggeration to say that the heatpipe is the figurative lifeblood of consumer computing today.

Desktop tower heatsink, with vertical heatpipes
Desktop tower heatsink, with vertical heatpipes

Laptop heeatsink, with two heatpipes
Laptop heatsink, with two heatpipes

What does a heatpipe do? Simply put, it transfers heat from one of its ends (the hotter end) to the other end (the cooler end), veeeeery quickly. It is one of the fastest feasible known ways for transferring heat, actually. And it is relatively cheap for its purpose. How does it work? I’d be happy to deliver a mini-lecture on thermal physics but this is not the place for it.

Laptop heatsinks? Yep, heatpipes there. Desktop towers? Yep, heatpipes there. Even the low-profile (1U) heatsink coolers have them (look at the bottom centre of the heatsink in the image below).

Gelid Slim Silence I-plus 1U cooler (with heatpipe)
Gelid Slim Silence I-plus 1U cooler (with heatpipe)

Anytime you want to move heat faster from one end to another, put in a heatpipe. If heatpipes are so great, why don’t we just use them to transfer heat directly to the surroundings?

Sure, why not?

Project Passive scratchbuild [HacknMod]
Project Passive scratchbuild [HacknMod]

This looks crazy, but scale it down to a lower TDP and you essentially have:

HDPlex H1.S Thin-ITX case
HDPlex H1.S Thin-ITX case. The heatpipes are secured to the side panels by flat aluminium plates, and to the CPU cooling block by a grooved aluminium top plate.

Heat from the CPU escapes directly to the two finned walls on the left and right. The fins increase the surface area for heat dissipation. This does not allow as much heat dissipation as a humongous tower heatsink with forced airflow, but it’s really good for something that requires no fans and is part of the case. But here we run into some special considerations.

Cooling non-CPU components #

Most active cooling (i.e. pump-assisted) setups have at least one fan moving air inside the case. This is important because while the other components (PCH, VRMs, etc) are designed to not require heatsink-assisted cooling, component designers do usually expect a minimal amount of airflow so that temperatures are not allowed to build up to potentially worrying levels.

Clearly this is not the case for Thin-ITX; while heat from the CPU is heatpiped directly to outside-facing surfaces, there is almost no airflow inside, so what can we do about that?

One solution is to use components specifically designed for lower power consumption (which usually correlates with lower heat dissipation as well). Commenter rrr lists some of them in his comment on my HDPlex H1.S review. There are SODIMMs designed to run at 1.35V instead of the typical 1.5V, which enables them to run cooler and use less power. There are also mSATA SSDs with low power consumption, particularly the Intel 525/530 series. As for the VRMs, not much can be done about them, but Intel’s Thin-ITX System Design Guide mentions that VRMs with more phases tend to run cooler because power is spread over more components (there are three VRMs on my Q87T motherboard).

As for the PCH, it is rumoured that the next generation of Intel processors, codenamed “Broadwell”, will have some models with the PCH integrated into the CPU package. This isn’t too surprising; Intel has already been making Haswell Ultrabook processors with such a layout.

Haswell-U Ultrabook CPU package. The smaller chip is the PCH.
Haswell-U Ultrabook CPU package. The smaller chip is the PCH.

Beyond Haswell #

What is potentially new with Broadwell is that this might make its way into higher-TDP processors for notebooks, and perhaps mini-desktops. My gut says it will probably not be available for desktops, but we will have to wait until we are close to Broadwell’s launch before these details will be busted/confirmed.

In any case, this would mean one more component gets to benefit from the higher cooling capacity of the case, which also means that fewer components are dissipating heat into the case. I wonder at which point we will see another allometric shift that changes the adjacent possible for passively cooled Thin-ITX systems and even smaller form-factors.

On to Thresholds in computing: Part 10 – Beyond Thin-ITX

See also

Thresholds in computing: Part 8 – Thin-ITX vs Mini-ITX