PDA

View Full Version : Noob question: relating CPU watts to temperature?

ski
12-01-2005, 06:34 PM
I'm doing a projection on CPU's of the future for a paper.

Right now, CPU's use about 150 Watts. In 2010 and beyond, they will use close to 200 Watts!

Now, from tinkering around, I know you don't want your CPU running at over 60 degrees Celsius, trying to aim for 40-45 degrees with just a Heatsink/Fan. So that's with a 150 Watt processor now.

Is there a direct way to see how much heat a 200W processor would generate, other than just projecting it to be 33% hotter since it's 33% more watts?

I can do Watts -> BTU's, but couldn't find anything for temperature :gle:

Showtime
12-01-2005, 06:42 PM
This might have some usefull info:
http://www.overclockers.com/tips30/
Also I remember a thread on it at AT, but not sure where it went...

Kevster
12-01-2005, 07:12 PM
Ski, just because a processor is consuming 200 watts of energy per hour doesn't necessarily correspond to how hot in temperature the processor will get. That is going to be a function of the radiated surface area of the processor, the thermal conductivity of the processor (die/casing), etc. When processors start approaching 200 watts of heat generation, there will still have a maximum operating temperature where the processor will fail. It is the job of the processor engineering and packaging teams to come up with better ways of getting rid of that heat load coming from the processor to maintain a recommended operating temperature. That becomes a steady-state heat transfer problem, once you reach the equilibrium point of heat generation rate to dissipation rate. You will be able to get a numeric value for a specific processor type depending on its' packaging and surface area, but to try and come up with a generic temperature value for the amount of energy consumed isn't going to work.

That aside, there is something that you're forgetting here: check your units.

Remember that power dissipation, or heat, is expressed in three different standard sets of units: watts, BTUs/hour and Calories per second. The watt is the standard unit of heat and power measure for electronic equipment for most countries around the world, including the US. A BTU is the amount of heat required to raise the temperature of 1 pound mass of water one degree Fahrenheit at standard pressure. A calorie is is the amount of heat required to raise the temperature of one gram of water one degree Celsius at standard pressure. You won't get a temperature value from this, only a heat load value that you have to dissipate somehow.

ski
12-01-2005, 07:45 PM
Wow, excellent info so far.

I guess I'll divulge some more specifics on the project. A lot of it is what Kevster touched on -- I'm looking at a roadmap for many aspects of CPU's to come in the future.

Page 22 of this document (http://www.itrs.net/Common/2004Update/2004Update.htm) (.pdf)

It lists that in 2008, the Allowable Maximum Power of a CPU under high-performance with a heatsink will be 198 Watts. It also gives power supply voltages.

So the basics say that here's a CPU with higher wattage, it's going to run hotter naturally. Basically, I see the need for better cooling with chips in the future -- just like how computers 5-10 years ago only needed heatsinks and no fans. I'm researching possible ways that engineers will come up with to cool these CPU's that have higher wattage.

Direct Immersion (in PFHC's)
"Water" cooling
Thermoelectrics

This class doesn't really go too deep into things, I'm an EE student so I only know the basics of thermo from my textbook, but it's not required to be in-depth in calculation -- simple back of the envelope computations.

So I'm trying to identify the problem as best I can (how much more cooling do you need for these higher wattage chips?) and looking at possible ways to handle the increased heat of next-gen CPU's.

InfiniteNothing
12-01-2005, 08:21 PM
Someone check my ricer math but I think you just look at the thermal resistance of the heat sink. If you heat sink has a thermal resistance of 0.5C/W (and I think we have to assume good convection) then you just have to multiply by the watts going through it. So if it's 150W your temp over ambient will be 75C. If ambient is 25C then you have a CPU at 100C. :)

To compare between different cooling systems just look at the different thermal resistances (obviously small is good right?)

ski
12-01-2005, 08:31 PM
Haha, I just got to the part of the article jel posted about the thermal resistance. Pretty simple stuff. :) Onward we go!