Original Link: https://www.anandtech.com/show/1296
CPU Heat Comparison: How Hot is Prescott?
by Derek Wilson on April 16, 2004 4:32 AM EST- Posted in
- CPUs
Introduction
There are plenty of issues with current methods of comparing temperature data between processors, and we have been hard at work trying to come up with something that gives us better and more comparable results than the temperature data reported by different processors' thermal diodes. In considering different approaches, we have run into the same problem over and over: how do we measure the total heat output of the processor in any nearly accurate way.The problem with thermal diodes is that between different architectures, there isn't really a way to compare what is being reported. The diodes only measure the temperature at a specific location, and the number we can pull out of a processor isn't necessarily representative of the average temperature of its surface.
Measuring the temperature increase of the system itself has issues as well. Different components in a computer heat up as well, and there's no way to really isolate the contribution of the processor itself to this process. The only number that can truly be compared between processors is the total energy output by the surface of the chip (or heat spreader).
The heat transfer from the CPU to the outside world would be a wonderful number to have, but there is the unfortunate necessity of a heatsink and fan for modern processors to run without cooking themselves. There isn't really a way for us to get to the surface of a chip in order to measure anything. So how are we supposed to measure how hot something is getting when we are required to be cooling it at the same time?
Well, there isn't any easy way we can think of, but just because something is impossible doesn't mean the AnandTech staff can't get it done.
Heatsink Modding: The New Rage
Our solution to the problem is admittedly a bit of a kludge. Nothing is perfect, but we really wanted numbers that were going to get us some useable data, so we needed to get as close to measuring the surface temperature of a processor as possible.We decided to drill a hole in a heatsink and drop in a thermister.
Our heatsink of choice was the Zalman CNPS7000A-AlCu. In order to get something consistent between processors, we wanted to measure the temperature at the center of the heatsink close to the surface that would contact the processor. We drilled in from the side of the block to the center of the copper strip. Since the retention mechanism was held in by a screw in the center of the side, we had to go in at an angle. Please do not try this at home, as flying metal shards and angled drilling into a block of metal are not things we want to be sued over should something go wrong. Here's a look at what we ended up with:
This is a close up of the side of our modded heatsink.
This heatsink fit on all the processors we wanted to test, and we were assured that degrading its cooling ability a little by drilling a hole in it would still allow it to keep the CPUs cool enough to run. We hooked up a thermistor that came with a case we had, and since it was already set up for displaying temps, we forewent building a reader for it ourselves and just ripped out the front panel as well.
This is the front panel display we used to read the temperatures.
Some care needs to be taken when reading the data that we collected using this heatsink. First of all, this isn't a measure of the heat output of a processor. This is a measurement of the temperature at a specific position in an open system. Having the misfortune of suffering through a thermodynamics class in college isn't always a bad thing (except with respect to one's GPA), but what this means is that these numbers still aren't perfect. The processor is heating the bottom of the heatsink, while the fan is cooling the rest of it. Even though the heat is being distributed through the heatsink, if there is a hot or cold spot very close to the position from which the thermister was reading temperatures, the data could be off for that particular processor. Because we are heating and cooling this object at the same time, the equilibrium temperature we find at this position may not be directly proportional to the heat output of the chip.
But, with all that said, this is still better than using the on die thermistor or monitoring change in system temperature. Our tests were performed in a temperature controlled room with no case on the computer. In order to load the processors, we ran two simultaneous instances of Prime95 for 40 minutes (though temperatures stabilized after about 25). Our idle temperatures were taken after powering up and doing nothing for 30 minutes (no power saving options were enabled).
Processor Temperature Comparisons
We can see that, at idle, Prescott is 6 degrees Celsius hotter than Northwood. Of course, its only 4 degrees hotter than the 3400+ (and cooler than what we are reading for the Athlon XP processors). It is important to note that the Athlon XP and Duron processors do not have heat spreaders on them, and therefore will dissipate all of their heat into a more concentrated area of the heatsink (directly below the thermistor). This could help account for their temperature readings, but since their interface with the heatsink is so different, it may be better to only compare them with eachother.
Under load the gap between Northwood and Prescott is cut down by a degree. These two are now the hottest processors we are looking at, but 5 degrees of difference isn't all that much in looking at the temperature at the bottom of the heatsink near the contact area.
In looking at temperature increase in percentage, there isn't much point in ordering the graph. We don't even know what's better here. A small increase could mean that your processor doesn't draw that much more power under load, or it could mean that your architecture is inefficient in saving power when idle. We thought that these numbers were interesting though, so we included them.
Final Words
From our data, it doesn't seem that Prescott is really that much hotter than Northwood. Like we mentioned earlier, though, heat output and our temperature measurements might not scale at the same rate. In other words, since Northwood is cooler than Prescott, our thermistor might be getting cooled even more by the fan. This could mean that Prescott and Northwood are even closer in total heat dissipation than in our temperature measurements. We are always working on ways to better collect this information, but hopefully what we have seen has been helpful.There is, of course, a temperature increase in Prescott though. But where did it come from? Prescott has about three times the number of transistors as Northwood (due to pipeline increases, the addition of 64bit functionality, and (not least) a doubling of the L2 cache). Prescott is fabbed on a 90 nanometer process rather than the 130 nanometer process of Northwood, which means that Prescott will have a higher power density.
There could also be some impact on increased temperature from Intel's new strained silicon technique. This increases the electron mobility through the body of a transistor. What this means is electrons move faster and transistors can switch on and off more quickly (something very good for high speed processors). Of course, this also means that transistors can end up leaking more current through them when they are off. This increases the power used by the chip which in turn increases heat output.
We asked Intel what (if any) effect actually using the 64bit extensions in Prescott would have on temperature, and we were told that it shouldn't have a significant impact on heat. Intel indicated that with the right 64bit application running we might see Prescott draw 2 or 3 more watts of power. Enabling and using the 64bit extensions will use parts of the chip that can currently remain happily disabled. Hopefully Intel will be right when they say that turning on this feature won't impact heat too much. Of course, we'll be there to test it out as soon as we can get ahold of a 64bit enabled chip.
We can't really be sure right now how much each of these factors affect Prescott's temperature, but all of them surely contribute.
The final issue we need to consider is the motherboard issue. Prescott is powered by a lower voltage than Northwood, but consumes more power. This means that it necessarily draws much more current. Though Intel did get the power requirements out to motherboard manufacturers, there may be some issues with Prescott support. Intel maintains that motherboards that were not designed for Prescott won't boot Prescott (and won't hurt either component), there sill may be some unforeseen issues, as even companies designing earlier P4 motherboards with an eye to Prescott wouldn't have had anything to test their motherboards with back when they shipped.
When it comes down to it, there are four options early P4 motherboards and Prescott. 1) Everything could work fine. 2) The system may not overclock very well. 3) The system may run but with reduced stability. 4) The system may not run at all. If there is enough interest, we may end up looking into Prescott and motherboard compatibility. Feel free to let us know if that would be something you would like to see.