Processor Efficiency

[Philippine Mango]

Processors these days run hot, so hot it's about 100W of heat dissipation. Now I've got a question, if the circuits with in a processor/the processor it's self were 100% efficient, wouldn't it not run hot at all? Is the reason why the CPU is running hot is because of the fact that it's running inefficiently? There is too much resistance within the circuits?

 

[f95toli]

No, efficiency has nothing to do with how hot a processor gets.
When you use energy you are in fact converting it to heat. The reason why you have 100W of heat dissipation is because you are using 100 J of energy per second.

The only way to reduce heat dissipation is to reduce the amount of energy used.

 

[Philippine Mango]

But isn't the heat generated by the fact there is resistance in the wires?

 

[f95toli]

Yes, but that is not the only source of heat.
Even if you would build a processor with superconducting wires that would only eliminate the dc-resistance. Much (most?) of the heat in a processor is generated when the transitors switch.

Heat is actually a serious problem even for superconducting electronics. However, a superconducting DSP circuit will still consume much less power than a similar circuit based on semiconductors.

There are also fundamental limits. Whenever you do work (e.g calculate something) you generate heat, that is one of the basic laws of thermodynamics so it is impossible to buld a processor that consumes no power.

 

[pm]

Besides which, in modern processors an increasing percentage of the energy lost to heat is through leakage due to the transistor not turning off completely. This problem is likely to get worse, or best case, stay about the same going forward. So regardless of the efficiency, the shear number of transistors leaking would create waste heat. It used to be that 99% of the heat produced by a CPU was from the transistors switching on and off, but recently it's more like 60% (ish) with the rest being lost through leakage.

 

[dmens]

From an architectural POV, it is difficult to define efficiency. Of course one can design a processor that is very efficient in terms of averaged power per uop but at great expense on performance. Without a TDP, crazy things can be done. I've seen about some strawman theoretical models that are real monsters... makes Prescott/Tejas look lean. Hitting the sweet spot on power/performance is the key issue at hand before a design team can proceed into development.

 

[TuxDave]

Originally posted by: Philippine Mango
Processors these days run hot, so hot it's about 100W of heat dissipation. Now I've got a question, if the circuits with in a processor/the processor it's self were 100% efficient, wouldn't it not run hot at all? Is the reason why the CPU is running hot is because of the fact that it's running inefficiently? There is too much resistance within the circuits?

There was an example of how energy is lost even assuming zero resistance when charge is moved from one place to another.

Imagine two equal sized capacitors, each with one side grounded and the other side attached to each other across an open switch. One capacitor is at V volts and contains C*V charge. The other is at 0 volts and holds 0 charge. Total energy stored in this system is C*V^2.

The switch closes and assuming no resistance, the C*V charge will distribute itself equally across both capacitors so each cap will be at V/2 volts. The total energy stored in this system is 2*(C*(V/2)^2) = 2*C*V^2/4 = 0.5*C*V^2. Where's the energy go? 🙂

 

[CTho9305]

There was an example of how energy is lost even assuming zero resistance when charge is moved from one place to another.

You might be interested in adiabatic switching - search google for some info about it.

 

[TuxDave]

Originally posted by: CTho9305

There was an example of how energy is lost even assuming zero resistance when charge is moved from one place to another.

You might be interested in adiabatic switching - search google for some info about it.

Yeah.. I studied that too. The only cost is speed...