Power and Heat

Even before we got these Duron chips, there was no doubt in our minds that the extra power and resultant extra heat coming from the AMD was going to be a major issue with the chips. We have seen nothing yet that has made us change our minds.

Joe Citarella’s article talks about the challenges and problems with air-cooling
the latest generation of AMD processors. If you think cooling a Duron is just like cooling a PIII or Celeron, you are in for a big surprise.

This article will try to give you an idea as to what the differences are between them and Intel chips, and to point out that things aren’t exactly cool on the Intel side, either, and are going to get hotter. The starting-off point for the data presented here comes from
the datasheets for the AMD Duron and T-Bird, along with the datasheets for the Celeron and PIII. However, most of the data
are extrapolations based on the data found in those datasheets and should be taken as my estimates rather than published references from Intel and/or AMD.

Here’s what you need to know about power and heat when you overclock:

Increasing voltage increases the wattage used by the chip much more than the percentage increase in wattage would lead you to believe.

An example:

The Celeron 533, 566, and 600 run at 1.5V, and take 11.2, 11.9 and 12.6 watts to run. Since each increase takes another .7 watts of power, one might suspect the Celeron 633 would take 13.3 watts.

However, the Celeron 633 runs at 1.65V, not 1.5V. You might expect the wattage of the Celeron to jump up another 10% as a result.

It doesn’t jump up 10% above the base 13.3 watts; it jumps up about 25% to 16.5 watts.

You see similiar wattage jumps from voltage increases with PIIIs and Athlons. Roughly, if you increase voltage X%, you increase wattage 2X%

This makes the published wattage estimates from Intel and AMD somewhat to mostly useless for overclocking purposes, especially for the Durons and Celerons. Looking at
wattage figures based on 1.5V is pretty misleading when you are running at 1.85V or more.

For overclocking purposes, there seems to be little if any difference between the wattage required by a low-end chip, and one required by a high-end chip.

The major difference between the AMD and Intel high-end and low-end chips is the amount of cache on the chip. Cache doesn’t chew up a lot of power. The old PII cache chips only required about 2 watts of power.

The relatively low wattage figures for the Celerons and Durons that are the prime overclocking candidates are being measured at relatively low speeds at 1.5V.

If you expect to run close to 1Ghz with either of these chips, 1.5V is not going to cut it.

When you add in the effects of both increased voltage and increased wattage, you get figures
very similiar to what you see for PIII or Thunderbirds. There might be a watt or two difference, but that’s meaningless for our purposes.

Even using the same voltage, you can’t extrapolate increases in wattage proportionate to the overall speed of the chip.

In other word, if a chip is running 50% faster, even at the same voltage, that doesn’t mean it will take 50% more power.

There is a linear relationship between Mhz and wattage, but it’s based on the specific Mhz increase, not overall increase at the same voltage.

For an Intel cB0 chip running at 1.65V, for every 50Mhz increase; the CPU needs about 1.7 more watts.

For AMD processors running at 1.7V, for every 50Mhz increase, you need about 2.2 extra watts. The increases are a little bigger or smaller if the voltage is higher or lower.

AMD chips are furnaces compared to Intel chips, but Intel chips are running into historically dangerous ground, too. The cC0 stepping (shrinkage) will bring little relief and in some ways make matters worse.

We have complete wattage estimates on the next page, but this is probably pretty representative of what you’ll face overclocking

Approximate wattage at 1.8V

AMD processors chew up about 25 watts more at a given speed and voltage than Intel chips, about 70-85% more power. That means 70-85% more heat than what Intel’s hottest chips ever are pumping out.

Generally, generations of Intel chips have topped out at around 30-35 watts, and the last
member of a clan has usually had some heat-related stability problems. We are running right
up against that boundary now with the cB0 stepping.

Intel has revised its datasheets to provide information on cC0 processors. cC0 processors will exist from 550Mhz and up.
They will all run at 1.70V, up from 1.65V. The only wattages shown are for the 1Ghz and 1.13Ghz processors. The 1Ghz processor shows power
consumption of 26.1 watts (as opposed to 33 for the current 1Ghz), but most of the reduction is due to it running at 1.7V as opposed to the 1.8V for the current model.
At 1.13Ghz (and 1.8V), power consumption goes up to 35.5 watts. cC0 is an improvement, and probably will make 1Ghz overclocking likely for air-coolers, but not much more than that.

Intel also has put up a new statistic which is even more troubling, something called “Power Density.” Basically, it’s how many watts are packed per square centimeter.
Chips are becoming smaller, but the power required to drive them at faster speeds isn’t dropping as fast, or isn’t dropping at all. You pump the same amount of power into a smaller space, and
it has to get hotter, and since there is less surface area from which heat can radiate, it’s harder to cool.

The Deschutes PIII 600 used 34.5 watts of power, and had a power density of 29.5 watts per square centimeter.

The cB0 Coppermine PIII 1Ghz uses 33 watts of power, a little less power than the Deschutes, but because it’s much smaller than the Descutes, it has a much higher power density of 45.5 watts per square centimeter.

The cC0 Coppermine PIII 1.13Ghz will use 35.5 watts of power, and since it uses a little more power and is a bit smaller than the cB0, it’s power density is 55.5 watts per square centimeter.

Just imagine what the power density of the Athlons must be like. Or what the first Willamettes will be like.

For now, for those of you contemplating Durons or T-Birds, if there’s a bit of a struggle keeping overclocked Intel chips cool with just air, what do you think adding 70-85% more heat on top of that is going to be like?

Don’t snicker, Intel folks, don’t expect the initial Willamettes to be much better. If you already have close to a space heater at 1Ghz, what’s 1.3 or 1.4Ghz or 1.5Ghz going to be like?

Athlon overclockers have had to deal with this for a while, so this is no news to them. However, I’m getting plenty of emails from people saying, “I’m going to buy a Duron/Thunderbird and put on a $15 fan/heatsink, that’s all I need.” Then I read, “I can’t reach 950Mhz. Why is this so hot?”

This is why you aren’t reaching 950Mhz and it’s so hot. You’re slamming 50 watts of heat into a cooling system that’s really meant to handle 25 watts, that’s why. If you’re thinking 1.2Ghz from a Peltier cooled T-Bird, you’re talking about 70-80 watts of heat, which will overwhelm most Peltiers out there today.

The next six months will be filled with literally hot chips. Temporary relief will only come with .13 micron shrinkage next year, but that only makes the surface area even smaller, and when the wattages creep up again as the chips ramp up in speed, cooling them will be harder than they are now.

Those who air-cool already have bricks to dissipate current heat. I don’t think we can go a lot further with just air.

It’s really hard to see how you’ll be able to overclock a lot a year from now with just air cooling. It’s just as hard to see how
anybody will be able to run processors, period, with just air cooling in two years.

Either cooling has to become more elaborate (and expensive); we go to multiple processors as a matter of course, or Intel/AMD start worrying about speed less and reducing power consumption more.

This is why we think you should be seriously considering at least watercooling for your next system, whether it be Intel or AMD. The AMDers are facing the heat now, but the initial Willamettes aren’t going to be a whole lot better.

Next page, wattage estimates.

Email Ed

How do I figure out my wattage?

If you like math:

1. Take the base wattage. For an Intel processor, use 20.8 watts for 800Mhz. For an AMD processor, use 42.6 watts for 800Mhz.
2. For every 50Mhz above 800Mhz, add 1.7 watts for Intel and 2.2 watts for AMD processors. Put this number on the side.
3. Take the base standard voltage for your processor. For Intel chips, use 1.65V. For AMD chips, use 1.7V.
4. Take the voltage you plan to run at, and divide that into the standard voltage.
5. Multiply the number you get by 2.
6. Take that number, subtract 1, and multiply the answer by the number you came up with in step 2.

Example:

Let’s say you want to know how many watts a Duron running at 950Mhz at 1.85V would chew up.

You take the base AMD figure at 800Mhz (42.6 watts). Since 950Mhz is 150 more than 800, you would add 3*2.2, or 6.6 watts to 42.6 watts, which gives you 49.2 watts.

Since you are running at higher voltage, you take your voltage 1.85V and divide it by the standard voltage (1.7). 1.85/1.7= 1.09.

Multiple 1.09 by 2, and you get 2.18.

Subtract 1, and you now have 1.18. Multiply that by 49.2 watts, and you come up with 58 watts. That’s roughly how much power your Duron will need to run, and how much heat you have to cool.

If you don’t like math:

Just look at the charts:

These are rough approximations based on little concrete data, and meant to be “one-size-fits-all.” That’s why you’ll see some anamolies in the numbers. A slightly more accurate model would be a lot more complicated. However, being off a watt or two
doesn’t matter for our purposes, so it’s good enough.

I’ve extended the range quite a bit beyond what you could reasonably expect from aircooling for the water and Peltier-cooling folks. The Peltier folks in particular need to look at this chart. If the CPU is putting out close to or more wattage than the rating on your Peltier, the Peltier
will do little, or even heat up the CPU.

XXXX indicates voltages below that recommended by the CPU manufacturer for their chip running at least at that speed.

Intel wattage estimates (cB0 Celeron/Coppermine):

Let’s now estimate Duron/T-Bird wattage.

AMD wattage charts (Duron/T-Bird)

Email Ed