An Interview with Arctic Silver
Or How Most Things I Knew About Cooling Were Wrong.
By Joel Hruska
Date: December 9, 2001
Earlier this week I had the privilege of interviewing Nevin House, owner of Arctic Silver, Inc. This is one piece that will test what you know about how processor temperature measurement takes place, what works keeping you CPU cool and what doesn’t.
===================================
VHJ: How did you end up in the thermal paste business? What gave you the idea for Arctic Silver?
Nevin: I began manufacturing and selling the Arctic Circle heatsink, which was a modified HP Turbocooler, about two years ago. The cooler the Arctic Circle was based upon was designed to handle a CPU that dissipated over one hundred watts of heat.
While selling the Arctic Circle, however, I became very frustrated with the quality of TIM (Thermal Interface Materials) available on the market. I began experimenting with various solutions in an attempt to generate a silver-based compound that would be electrically non-conductive. One of the things I’ve learned working in this business, and something a lot of people have trouble with, is how important it is to think on a microscopic level about what is really happening between heatsink and CPU. Arctic Silver was designed with these precepts firmly in mind.
VHJ: Why use silver?
Nevin: Silver is the most conductive of all the commonly available metals and is relatively willing to enter a state of fluidic suspension. Although copper is another choice, it does not enter a suspended state as readily and is harder to work with. On a microscopic level, the molecules of Arctic Silver are arranged in a plate or poker-chip shape and rest between the heatsink and the CPU itself.
VHJ: What typical improvements do customers see when switching to Arctic Silver II?
Nevin: The average temperature drop from switching to AS II from a standard heatsink compound is between 2-7 degrees Celsius. Obviously this is going to vary depending on the heatsink used and the type of CPU, but temperature drops of these amounts are considered average.
VHJ: How do you explain the fact that many websites do not show nearly this much of a temperature drop when different types of thermal paste are tested? Oftentimes thermal paste roundups show differences of only 1-2 degrees between all the pastes themselves and this variation well within a margin of error.
Nevin: You raise an excellent, excellent point and one I wanted to address. Let’s look at a bit of history:
When Arctic Silver first arrived, it was tested by a variety of websites and yielded results showing temperature drops of 2-7 degrees C -- exactly what we predict. The testbeds used by these websites, however, were primarily Pentium III’s with internal temperature monitoring thermal diodes.
Today, most websites have switched to Athlon-based systems which use a thermistor in the socket for temperature measurement. Unfortunately, this is inherently flawed. According to basic laws of physics, it is impossible to obtain an accurate temperature measurement downstream from the heat source in a secondary heat path in a non-isothermal environment.
Allow me to illustrate further. Imagine a room with one outside window. It’s winter outside—so the temperature outside is 30 degrees, while the room’s temperature is a uniform 70 degrees. A thermistor placed inside the room will, of course, register a temperature of 70’.
Now, imagine what would happen if you took that same thermistor and taped it to the outside of the window. The thermistor will register a MUCH lower temperature than 70’. There is only one section of it even in contact to receive heat flowing from the room, and that heat must pass through the glass. Furthermore, the other 75% of the thermistor is in constant contact with the outside air—which, as we have noted is much colder than the air in the room.
In short, the in-socket thermistor cannot effectively measure the temperature of the processor, all it truly does is measure the temperature of itself.
VHJ: Even if what you say is true, can’t an external thermistor still be accurate, but low? For example, if a core is running 15 degrees Celsius hotter than the thermistor reports, can’t we assume that a 3 degree drop in reported temp equals a 3 degree drop in actual temp?
Nevin: No, you can’t, but I’ll explain why. Again, let me use an example: Imagine you have a stove turned to "Bake" and running at 400° F, will the top of the stove be 400°?
Of course not. Now, imagine turning up the heat to 500°. Will the temperature on the top of the stove rise 100°?
Once again, no. But why not? The reason is because of something called thermal compression. Thermal compression occurs because intervening materials between the source of heat and its heat path offer resistance and block some of the heat from being transmitted. A 100° increase in the heat of the oven may only translate into a 15-30 degree increase on the top of the stove.
Now, how does this relate to temperature measurement? As a CPU heats up, the temperature does increase, but because the thermistor is external, it is subject to a high level of thermal compression. As CPU internal core temperature rises so does the thermistor temperature, but subject to the problem of thermal compression.
Just like our stove example, where a 100° temperature rise internally left only a 15-30° measurable rise externally, a 15-20° Celsius rise internally in a CPU may leave only a 3-4° rise externally. Because all Socket A solutions save for one of Siemen’s motherboards for the Athlon XP currently use external measurements it means that all Socket A solutions are wrong [ed: at least one Compaq system uses the onboard XP diode]. To provide measurements that are at least in the ballpark, current motherboards add 10°C to 15°C to the measurements they receive from the thermistor, but this fixed correction does nothing to eliminate the compression error.
VHJ: It sounds as though you’re saying that almost all heatsink and thermal paste round-ups are wrong and inaccurate.
Nevin: Frankly, most of them are, and its very frustrating. Take a look at a review of two or three high-end coolers sometime, and if you watch, you’ll notice that at really high levels, two coolers that have been performing within, say, two-to-three degrees of each other all night will suddenly have a dramatic difference with one cooler able to complete a test, and one failing. The reason is, there was not a two-to-three degree difference; once thermal compression is factored in, the actual difference becomes an 8-12 degrees in Celsius. That’s quite a jump.
The thermal paste round-ups suffer from similar problems. On the Arctic Silver website it says to give the paste forty-eight hours to achieve maximum effective coverage. I’ve tested this extensively and watched temperatures on my own test equipment drop 2 ½ to 3 degrees Celsius due solely to this process. Most reviewers slap the stuff on, wait an hour, and take their measurement, but they aren’t using the product as its meant to be used when they do.
VHJ: Its been suggested on a few sites that as CPU dies shrink the need for specialized thermal compound is much smaller because the die area for the compound to work in is much smaller. [ed: Huh? Die shrinks increase thermal density, making heat dissipation more difficult.] Is this true? Is Arctic Silver being driven out of the market by shrinking processors?
Nevin: No. Actually, this is completely backwards. As computer dies shrink the need for a high-quality TIM increases enormously. Imagine two theoretical TIM’s: one with a thermal resistance of .02 and one with a thermal resistance of .002 for a 1 square inch contact area. Because the second compound has a thermal resistance 1/10 as high as the first compound, any shift in temperature which would cause a 1 degree rise in temperature using the first compound will only cause a 0.1° rise in temperature using the second (we are not factoring in heatsink resistance in order to keep our example simple).
Imagine, however, that the die we are working with shrinks to 1/6 its former size (T-Bird size). The same amount of heat must now escape through a MUCH smaller area, meaning that the rise in thermal resistance (and hence, in temperature) will be much greater. Looking back at our original numbers, the same compound that caused a rise in temperature of 1° C, applied to a die 1/6 the size, will cause a rise in temperature of 6° C. Our compound with a thermal resistance 10 times better than the first, however, causes a rise of 0.6° C.
So where we had a difference of less than 1° C on a 1 square inch contact area, we have a difference of 5.4° C on a contact area 1/6th as large.
The reason Arctic Silver is looking less effective on AMD systems using external temperature measurements, in short, is caused by the natural measurement compression that all external measurements are subject to. Most heatsink or thermal paste round-ups show a performance difference today of 2-3 degrees between thermal pastes and 4-8 degrees between heatsinks. Because A.S. II gives a benefit of 2-7 degrees Celsius it is easily masked by a high compression factor in the 3 to 1 or 4 to 1 range.
VHJ: Obviously you see inaccurate thermal measurements as a huge problem. Do you believe it’s a problem that will be solved in the immediate future?
Nevin: It’s a problem that’s easily solvable once more Athlon XP motherboards ship that support internal temperature monitoring. Right now there’s only one Siemen’s motherboard that supports the internal thermal diode on the Athlon XP -- that’s something that should be changing soon [ed: we mention again the Compaq system]. Once it does, we’ll see much more realistic temperature measurements. The idea that you can get a proper measurement from any type of external thermistor is simply laughable. No matter how close you get to the die, no matter where you position the probe, you are still taking a measurement subject to thermal compression and still receiving bad temperature data as the probe is surrounded by air on all sides of a different temperature than the CPU core itself. The only accurate way to measure temperature in a CPU is using an internal thermal solution such as those supported on the P3, P4, and Athlon XP.
VHJ: You’ve recently released a new product called Arctic Alumina. How is it positioned in the market and what demand do you see it filling?
Nevin: Arctic Alumina is a more ‘budget’ oriented thermal paste that is highly competitive in its class and price range. It was developed for the ‘casual’ computer builder who doesn’t want to spend $9 on a little tube of paste that, from his perspective, he’ll only use once. With its lower price point, Arctic Alumina is a product designed to appeal to that buyer and give him a first-rate product for only about 1/3 the cost.
VHJ: Does Arctic Silver have any plans to begin shipping product with major heatsink vendors?
Nevin: At this time we have no hard-and-firm contract deals with any heatsink vendors. However, with the growth of performance heatsinks and the interest of the DIY market in these units we believe there’s tremendous potential for this type of deal between our company and a high-performance manufacturer. We remain open to the idea and are exploring our options.
===================================
VHJ would like to thank Nevin House for his time, insights, and extremely thought provoking conversation.
===================================
Pssst! Our Shopping Page has been updated.
===================================