|
|
Thermal specificationsDisclaimer: This data has been reproduced from various Intel datasheets. I am not responsible for the accuracy of this data or any typos. Plus not all the data is here yet!!!One of the most significant barriers to designing and operating a stable microprocessor is heat. Virtually anything you can do to make your computer run faster or more stable involves generating more heat. Overclocking your computer generates more heat. Increasing the supply voltage generates more heat. If the chip heats up too much, the characteristics of the devices inside the chip will change enough so that the microprocessor will fail. Obviously, this is not a good thing! What I've decided to do here is place the thermal specifications for Intel's microprocessors on the web and give just a little bit of interpretation (they do pretty much speak for themselves). Measuring the temperature of the chip Intel specifies an allowable operating range of temperature for their microprocessors. The value they specify is Tc, which stands for case temperature. The case temperature is the temperature directly at the top of the chip packaging. Clearly Intel doesn't specify the temperature of the microprocessor inside the packaging itself, which would be specified as Tj (for junction temperature), because you can't measure that in any practical way. Typically the operating temperature range for the microprocessor is between 0-70º C, although sometimes the specification varies for different lines of chips. The Pentium Pro, for example, has a specified operating temperature range of Tc = 0-85º C. All of Intel's measurements and data go by measuring the case temperature. The procedure they use to do this follows: The device used to measure the temperature of the case is known as a thermocouple. In the case of measuring the temperature of the microprocessor the thermocouple to be used looks like a wire with a little beaded metal end exposed. The little bead at the end is actually the junction between two wires made of dissimilar metals. An excellent primer on thermocouples and the different types of metals commonly used to make them can be found here. Intel specifies 35-gauge or finer K, T or J type thermocouples. They recommend attaching the bead to the case packaging with some type of thermally conductive material. I don't think that is really crucial for our purposes, but whatever. They also say to contact the case with the thermocouple at a 90 degree angle...that is, at an angle normal to the plane of the casing. Anyhow, it is self-explanatory if you want to measure the temperature of the chip without a heatsink. If you want to do it with a heatsink or HS/fan combo attached, Intel specifies that you must drill a hole in the heat sink no more than 0.15" in diameter and go through it to contact the chip case. If you are lucky enough to have a board like the ASUS TX-97 series, you have it much easier! These boards have the built-in capability to measure chip and motherboard/ambient temperature. All you should need to do is go into your BIOS to see just how hot your processor is getting. Anyhow, we don't have to be too accurate...a couple of degrees off here or there is not that important since we know one thing...the lower the better!!! A general idea of just how hot your processor is will suffice. BTW, the key equation you want to know here is: TA = TC - (P*ThetaCA) TA = Ambient temperature Case temperature is the variable you want to keep as low as possible, so obviously, you can do this by modfiying the other three variables. TA, the ambient temperature, can be modified in a number of different ways. You could turn up the airconditioning in your house (like those really cold computer rooms in big companies). Or you could place some fans in your system to circulate the cooler outside air into the case and make sure that hot air doesn't stagnate inside. Or you could remove the case altogether if that doesn't bother you! In case you don't know this, room temperature is usually taken to be around 27 º. Obviously the real room temperature depends on where your computer is. P, the maximum power consumption, can only be modified by changing the voltage or speed on the chip. Since the whole purpose of cooling your chip is so you don't have to clock your chip down or other similar stuff to keep it stable, I think we can safely ignore that. ThetaCA, case-to-ambient thermal resistance, is a measurement of the effectiveness of your heatsink/fan. The lower this figure is, the better.
Thermal Resistances for SPGA Packages with Heat Spreader
Thermal Resistances for SPGA Packages without Heat Spreader - Pentium Processor 75, 90, 100, 120 MHz
Having a heat spreader on top of the chip (metal plate thingy) is clearly a good thing. Pentium Classic - Power Dissipation Requirements for Thermal Design
Hmm. Well, first, if you take a look at the other chips in this section, notably the Pentium Pro and Pentium II, you will see that the Pentium Classic chips really don't dissipate much power at all. In other words, they are fairly easy to keep cool. You might also notice that the power dissipation increases from 75-120 MHz and then drops starting with 133 and increases again up to 200. This is because the 133-200 MHz models are produced on a 0.35 micron process, which not only shrinks the size of the chip but allows the design to draw less power and therefore produce less heat. Pentium MMX - Power Dissipation Requirements for Thermal Design
AMD K6 Typical and Maximum Power Dissipation
Well, as you might expect, the AMD K6 runs significantly hotter than Intel's Pentium or Pentium MMX lines, so you need to pay much more attention to cooling when dealing with the K6. You can't get away with low-profile low quality heatsinks like you might be able to with a Pentium, especially if you overclock. Note in particular the dramatic increase in maximum power dissipation when moving from the 200 MHz part to the 233 MHz part. In case you don't know this, power dissipation scales linearly with speed/frequency but scales with the square of the voltage. So take a good look at the Pmax figure of the 200 MHz K6, 20 W. Divide 20 by 200 and multiply by 233 to get the Pmax of the 200 MHz K6 if you were to run it at 233 MHz at 2.9V. (20 W/200 MHz) * 233 MHz = 23.3 W Now factor in the increase in voltage by dividing by the square of 2.9V and multiplying by the square of 3.2V. (23.3 W/2.92) * 3.22 = 28.37 W Well, look at that! It's the same Pmax as the 233 MHz K6! If you are familiar with overclocking, you know that the increase in voltage to 3.2V is probably needed to keep the chip stable (at the expense of producing more heat/power). Now you know how AMD "overclocks" their CPU's! :-)
Cyrix 6x86MX - Power Dissipation
You can see that the 6x86MX chips have to dissipate just about as much power as the K6's do. Similar strategies for cooling need to be taken.
Pentium Pro - Thermal Design Power
Wow! Look at how much power these guys have to dissipate! Pentiums have it easy, don't you think? All Pentium Pro's come with heat spreaders and because of the dual cavity design of the processor (one half chip, other half cache) Intel actually tells you to measure the heat over the center of the processor core, not the physical center of the heat spreader/packaging. You can clearly see that the cache does play a role in total power dissipation however. Another interesting tidbit of information, the 256K L2 cache is manufactured on a 0.6 micron process; the 512K L2, on a 0.35 micron process. Pentium Pro - Case to Ambient Thermal Resistance
Pentium Pro - Ambient Temperature required per Heat Sink Height for 29.2W and 85º Case
Pentium Pro - Ambient Temperature required per Heat Sink Height for 40W and 85º Case
Now this is some really great data and something us home users can put to use. The 29.2W figure corresponds to the maximum power dissipation of the 150 MHz PPro and the 40W figure roughly corresponds to the 200 MHz 512K cache PPro...actually a little above that, so we might consider this specification something of an upper limit. First of all, the data tells you what you already know, bigger is better. Bigger fans and bigger heatsinks keep the chip cooler. No big surprise, right? But look at the significant difference in performance between 0.5 to 1.0" height fins and even from 1.0 to 1.5"! In the case where you need to dissipate 40W it can mean a difference of 20º or 30 º C to move from 0.5" to 1.0" fins, and even from 1.0" to 1.5" buys you an extra 6-8º. In most cases, 2.0" fins seem like overkill. And anything near 0.5" fin height looks outright terrible. You might note that the fabled PC Power & Cooling fans are of rather inadequate length, plus the fans are inside the heatsink itself, further reducing the surface area!!! This is why, in my opinion, PC Power & Cooling PPro fans are mostly marketing hype and not a serious solution for cooling your CPU. Which brings to mind another point. Ever hear of those "low-profile" heatsinks? Make sure you understand this, the words "low-profile" and "heatsink" don't mix. The next thing to note is that adequate airflow in the neighborhood of around 400 LFM is requisite, otherwise the effectiveness of your heatsink will drop off dramatically. Doubling your airflow to 800 LFM can get you several degrees cooler as well, so don't underestimate the importance of slapping a big fan on your heatsink or using a secondary case fan to blow over the CPU. I'd say your typical heatsink might have 1.0" fins and a 400 LFM fan (I think this roughly corresponds to most 40mm fans). Increasing the fin height to 1.5" and doubling the LFM to 800 buys you an extra 15 degree increase in Ta to maintain the Tc at 85 º! Not bad...and some people don't even have heatsinks that are 1.0" tall?! BTW, Intel's data is based on keeping the Tc of the Pentium Pro at 85º. I'd personally try to stay well below that, especially if you are overclocking. Heat is the Pentium Pro's primary enemy...just look at how much power it has to get rid of! I think a reasonable goal might be to keep the processor in the range of 55-65 º, but that is just speculation on my part right now. I can give you a little data of my own, though. My system consists of two PPro 166 w/512K cache o/c'ed to 210 MHz, each with 1.0" pin fin heatsink (width and length roughly the size of the CPU) and 40mm ball bearing fans. Ambient temperature between my CPU's hovers at around 41 º. Measuring the edge of the heat spreader between the sink and CPU gives me a reading of around 50.5 º. (No, I didn't drill holes in my heatsinks.) I measured the temperature using a DM23XT from WaveTek with the included Type K beaded thermocouple wire. With the case on, I expect the case temperature to rise significantly...I'm currently searching for a way to improve airflow significantly through the case. I'm expecting to buy some new and very large heatsinks with large fans (60mm) as well. And so should you! Update 1/4/98: After a while playing around with my case and the airflow inside, I have to emphasize that in order to properly cool your CPU you cannot ignore the airflow inside your case. This is critical to cooling your CPU...equally as important as the size of the heatsink itself. Make sure fresh air is being pushed into the case in order to avoid recirculating hot air. Try to have an exhaust fan to pull out the hot air. Make sure the airflow makes sense, e.g. don't put two fans in the front of your case, one blowing in and the other blowing out. Finally, if you can rig an 80mm case fan to blow over your CPU's, that can go a long way towards keeping your CPU cool.
Pentium II Thermal Design Specifications
Little note here: TPLATE is basically the equivalent of TC (TCASE). The Pentium II dissipates less power even though it runs at a faster clock speed than the Pentium Pro because its cache only runs at half the processor frequency and (the biggest reason here) its core voltage is 2.8V as opposed to 3.3V for most Pentium Pro's. Update 1/28/98: Check out the new 333 MHz Pentium II! You are probably thinking, wow, that sucker stays cool! That's because the new Deschutes line of Pentium II (333 MHz and above as of now) are manufactured on a 0.25 micron process instead of the 0.35 micron process. That lets Intel run the Deschutes 333 MHz Pentium II down at a Vcore of 2.0V as opposed to 2.8V. The power dissipation of the 333 MHz Pentium II is almost half of the 300 MHz PII! Example Thermal Solution Performance for 266 MHz Pentium II Processor at Thermal Plate Power of 37.0 W
DatasheetsPentium datasheet Pentium MMX datasheet Pentium Pro datasheet Pentium II datasheet Thermal Management: Boxed Intel Pentium(R) Pro Processor (online HTML) AP-525 Pentium(R) Pro Processor Thermal Design Guidelines Cyrix Hardware Design Center
(lots of assorted documents here) AMD K6 MMX Technical Documentation (lots of
assorted documents) AAVID Thermal Technologies - technical papers (very good) Some Typical Overclocking Ranges These are the typical ranges I've seen others report successfully reaching. Obviously there is no cut-and-dried definition for what you can and cannot get, but it should give you a general idea of what your chances are and what you can reasonably expect. "Usually reaches" means almost everybody can o/c to this speed. "Sometimes reaches" means that some people can get to this speed, but if you happen to make it you should consider yourself lucky.
Richard Kuo, | ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
