Common Cooling Misconceptions

A collection of corrections – Johnathon Smith

Large car radiators have a higher flow resistance

Actually, with a full size car radiator, it is easier to get a high flow rate. There are more parallel paths for the water to travel. The flow in each tube may be less, but overall you get a net greater flow rate. Most heater core style designs have the water go down into a collection tank on the end and then return up to the top, so the water travels twice the length of the heater core.

For a heater core with 2 120 mm fans to fit on it, the water must go 4x that length in only maybe 6 tubes each way. Most radiators you will find are not going to have total path lengths longer than that (because typically the water only crosses once to another tank then out), yet they may have 20+ parallel tubes.

Now you tell me which one will have the least resistance to flow. I think the little thought experiment is sufficient and from testing on my own setup, I find a good flow rate increase by changing to a full size radiator. Still think it is hard to get high flow rates with big radiatorors? I hope not.

Why cold plates are used and mounting pressure

The downside to cold plates is with adding materials that have a good thermal conductivity, you are adding distance, and more distance gives some thermal gradient over the distance it is carried. The thermal gradient, all else being equal, would mean the block parts transferring heat to the water have a smaller temperature difference and thus would transfer less heat given the same hot object temperature. In practice the hot object just runs hotter and cooling is worse.

The gain you get is a more even heat distribution over the whole base of the water block (or whatever heat sink in use), so if your water block was designed to have the heat going into the water over a large portion of the base, and if the base was sufficiently worse that the detrimental effects of the added distance is overcome through better spread of the thermal density into the base of the block, then it will do better with the cold plate.

The other gain that cold plates get you is the ability to mount objects like peltiers with much more clamping force than is otherwise possible. More mounting pressure helps lessen the temperature drops across the thermal interfaces.

Peltier performance and heat output

Have you ever wondered just how well a peltier would do on your video card or CPU? If you are afraid of math you might want to skip this.

To begin with we shall use an idealized model for how the cooling system will work (and one that agrees fairly well with my results).

A highly overclocked CPU may push 100 ~ 150 watts at full load – that is the initial heat source. A typical room temperature of 20C will be assumed. Reasonable cooling for a water block on a peltier would be around 0.10 C/W. A typical water pump will add about 10~20 watts to the water.

My general conclusion is that most people do not know what any of the numbers mean, much less what to do with them to get some useable information on what size peltier would be good for them. Taking a simple 226 watt unit, at full power it will use 24 amps * 15.2v = 364.8 watts. That is a rather large amount of heat to cool off in itself, but you have to add to that the heat that the CPU makes and you get something hovering near the 450~500w mark.

If your water block is only good for .10 C/W then you have a 45~50C temperature loss between the hot side of the peltier and the water temperature. Given then a typical 226 watt unit has a max delta T of about 67C, then, as you can see, most of the sub-ambient temperature capability is already lost.

If the CPU only makes 100 watts, then the temperature across the peltier
is 67(C) * (1-100/226) = 37C or so.

that kind of heat, so you end up with the CPU running around 10C – 37C +
45C = 18C at full power over ambient air temperature. That is bad.

A similarly equipped setup without the peltier would yield something more like 13C over ambient. Granted this is an oversimplified case and there are more places for temperature gradients with the use of a peltier and with a good enough water block and high enough flow rate you can get the C/W down really low with such large objects since thermal density is not too extreme, but it does take foresight to do it right. It is also harder to get really low C/W values on smaller CPU cores, but I hope my oversimplified example shows how it is getting towards impractical to use such devices on modern CPUs.

Many users seem to think that putting an 80 watt unit @ 12v on a GPU is a good idea – I mean after all, it worked good on the old ti4200s. Xbitlabs.com has some really cool data on actual power usage of modern video cards though. As can be seen HERE and HERE, a 6800ultra at stock speeds can use over 72 watts. Not all of that is being used by the core, of course, but I would wager that a lot of it is. That is hard to cool with an 80 watt peltier.
{mospagebreak}

Water after a CPU and/or GPU block is too hot to cool something else

Think about how much heat the CPU and GPU make and are putting into the water. Then consider how fast the water is cycling through your system. How much hotter is the water after passing through the CPU and GPU blocks than when it started?

A quick little calculation using the atrocious values of 1 Liter/minute and 200 watts total heat going in with CPU and GPU combined will yield that (whips out calculator in Windows) the water would be about 2.9C hotter by the time it hits the hard drive than when it hit the first CPU block. Given that your flow rate may be as high as 4L/minute and the heat being put in the circuit is probably less, you will have no problem adding a hard drive block in at the end of the loop. You can always put it first if a couple degrees are that big of a deal for you.

As a side note, I have noticed that some people seem to think that if they are cooling something more with the same cooling, like cooling a CPU more without changing anything but maybe removing an elbow in their system or whatever, that sometimes people think that the water will get hotter.

I am not sure how this myth came about, but if you simply remember the whole “energy is conserved” thing that was said in your basic science class, then you can think it through to see that, in fact, the water remains at the same temperature. How much heat does the CPU make if it is a little hotter or colder? For our purposes it is relatively the same, though I am guessing that it does change somewhat.

If the same amount of heat, which is just thermal energy, is being put into the water, then how hot will the water get? Does the water “care” how hot the device making the heat is? It does not. The water will go to the same temperature, all else being equal.

Error in temperature measurements

I have seen people saying that the onboard measurements of CPU temperature are too high by 5~10 C because of some thermal probe they taped to the side of an IHS or water block or some other similar method. I do not deny that the onboard temperature sensors do give erroneous readings more often than not, and sometimes are quite a bit off, but the method of measuring a temperature with a probe that has half of itself exposed to air and having the probe a distance away from the core is assured of giving temperature readings too low.

Copper thermal conduction and aluminum better heat ‘release’ or some such nonsense

Some people say copper cannot dissipate heat as well as aluminum, but say that copper absorbs heat better and that copper transfers heat across itself better than aluminum. It is even implied that it transfers heat to aluminum well, just not very well to air. So this implies that copper transfers heat to itself, CPUs, aluminum, heat spreaders, peltiers, silver, water, and who knows what else better than aluminum, just not better to air. The whole argument is nonsensical to begin with if one thinks it through.

The rate of heat exchange is based upon the difference in temperature, with all else being equal, between the metals. Copper conducts better so it has a higher temperature difference, so it dissipates heat better than aluminum.

As a kind of counter situation which some may be trying to prod at though, the specific heat of copper is .0923 cal/gK and it has a density of 8.92g/mL. Aluminum has a specific heat of .215 cal/gK and a density of 2.702 g/mL (Chemfinder.com).

If you take those and figure out the heat capacity per volume, then you get 0.82 cal/mL*K copper and 0.58 cal/mL*K for aluminum. It seems like some unnamed computer sites may think that their equivalent copper heat sink takes longer to cool down than the aluminum one because per volume, the copper will store more thermal energy per temperature. That is very true, but it does not deal with how well the metals conduct heat or anything else that is often implied.

Radiation vs. forced airflow and which material is best suited to which

Have you ever been next to a campfire and felt the heat from a distance? That was thermal radiation.

In general, for computer cooling, radiation is a relatively small part of the total cooling process. Most of the cooling is done by having some fan force air over metal fins and then the air caries away the heat. I often see people saying that aluminum is best for ram sinks and low airflow fan heat sinks while they say copper is best for “pure performance” or whatever that means.

In truth, copper would give better thermal performance in all cases, but it costs more and weighs significantly more. Hanging a pound of copper from 4 holes around a CPU may not be a big deal, but it is not advisable to glue that to a RAM chip 🙂 With lower amounts of heat and less thermal density on objects like RAM chips and mosfets than, say, a CPU, using aluminum has nearly the same cooling performance as copper but it is lighter and cheaper.
{mospagebreak}

Heat output changes with frequency and voltage

With only a little browsing around some forums, it becomes apparent that most users are not fully aware of how the heat output, and thus cooling challenge, changes due to more overclocking.

In the past on 130 nm and larger process made chips, the heat was largely linear with clock speed and went up as a square of the voltage increase. That means that if you double clock speed then you double the heat the chip makes. If you double the voltage though, you quadruple the heat. A somewhat conservative over voltage of 15% then gives about 32% more heat, whereas increasing the speed of your chip from 2 GHz to 2.5 GHz is only 25% more heat. If you did both however, then it is more like 65% more heat than original.

That was with the older designed chips. Given that Intel and AMD CPUs are similar to IBM’s new 90 nm process, we can use the data from IBM CPUs and it can be seen that now increasing voltage gives an exponential increase in heat. Raising the voltage makes cooling challenges extremely harder than otherwise.

CPU speed scaling with temperature changes

I have only done testing with a Pentium 4 Northwood core and, to lesser extent, with a Thoroughbred A 1700+ chip. In short, many people seem to overestimate the gains from losing a few degrees Celsius.

On the Pentium 4 chip, testing conditions were 1.7v with an Abit IC7 v1.1 (and a v1.0) motherboard and a Turbocool 475 power supply, all running a 2.4C from Costa Rica. RAM was kept at 2/3 ratio and 3 brands were used over the course of time. The actual CPU temperatures are unknown, but the water temperature varied from 40F up to 100F. Any higher voltage resulted in no net gains at any temperature.

The later Northwoods seemed to mostly have a cutoff voltage beyond which instability ensued regardless. At 100F, water temperature the stable speed was 3450 MHz. With 40F water, the maximum speed attainable stable was 3540 MHz. I doubt that the chipset was the limiting factor, as a voltage mod was done and no speed gains were observed.

Cooling was also increased on the motherboard with no gains as well – a 90 MHz increase over 60F water temperature change is hardly earth shattering. Converting to the more commonly used Celsius units, that gives something like 2.7 MHz/C scaling, which appeared fairly linear.

In contrast to that was the Thoroughbred A chip which was rather sensitive to even the room heater coming on for many more MHz, but I did not take good measurements to quantify the results.

The conclusion is that not all CPUs scale similarly with temperature, with some seeing very little gain. Given the large starting speeds that modern CPUs have and the low relative gain in MHz/C (about 1/1000 of operating speed per degree), it is clear that very large temperature changes are the only ones that make sense when investing in new cooling equipment, but by the same token not all CPUs scale the same.

The other tidbit of interest to gain from this is that to get full gains from new cooling gear, increased voltage is almost required if the temperature scaling is severely poor (as in the case of several late made Northwood cores).

Condensation due to AC blowing in a case or having a case outside in the cold

There is a reason that every time it gets cold outside everything does not become soaking wet, and a reason that it rains and things get wet. Rain happens because the water vapor content in air becomes greater than what the air can maintain, so it condenses out. A similar thing happens when you take a Coke out of the fridge and it gets wet with condensation – the air around it was warmer and had greater capacity to hold water vapor, but then it got cold and the water condensed out. This has to do with the vapor pressure of water and the temperature dependence.

The important thing to realize in the first two cases is why the water condensed out – it was there to begin with. When you put a case outside, there is no excess water to begin with, so nothing happens until you bring it inside. With an air conditioner recycling air through a case, the air conditioner has cold coils which cools the air and the water condense onto them.

If you have any questions, comments, improvements or anything, please email me.