What exactly happens when we overclock a processor?

[abhaybhegde]

Hi, was wondering what happens when a processor is overclocked? I am not asking in terms of the FSB/QPI stuff..but rather the basic concept.

Are we actually increasing the frequency?
How are power and frequency related?(if any)

Regards,
Abhay

 

[ShawnD1]

Electricity flows every time the cpu pulses. Pulsing more times means more electricity. Increasing voltage generates more heat for each pulse.

 

[Daedalus685]

Are we actually increasing the frequency?

The frequency is increasing, which gives faster performance. In order to get the frequency higher we must raise the voltage, which is propotional to the state change time of the transistors. The higher the voltage the faster the switches can change state, and the less likely errors are when upping the frequency.

How are power and frequency related?(if any)

Power is related to the voltage draw and the frequency. Raising either increases the heat produced and the power consumed by the CPU. Roughly, the power is equal to (C)(f)(V^2) where V is the voltage, C is the capacitance, and f is the switching frequency.


What you are doing with overclocking is trying to increase the frequency. To achieve that you must increase the voltage, the cost is always power consumption and heat.

Electricity flows every time the cpu pulses. Pulsing more times means more electricity. Increasing voltage generates more heat for each pulse.

That is such an over simplification that it borders on incorrect. The electricity is not pulsing, the states of the switches are what is changing. The electricity is always flowing through the circuit and used to change the states. Voltage and frequency are not related to each other in that sense. Increasing voltage increases power draw as in all electronics, but increasing frequency does as well. They are independent variables in this situation, we relate voltage so closely to frequency because voltage determines what is the real world maintainable frequency. (EDIT: Given I think I understand what you were getting at it doesn't seem like such a poor simplification to me any more )

 

[The Sauce]

Think of your CPU as analogous to a water wheel. How fast the wheel spins is the frequency. How much water flow it receives is the voltage. To make the wheel spin faster you must increase the amount of water flow. Then the wheel can do more work (i.e. generate more power, etc.). Eventually, adding more water flow will not speed up the wheel any more and will just become dangerous and possibly damage the wheel. Well, thats the best I can think of.

 

[Zap]

 

[pm]

That is such an over simplification that it borders on incorrect. The electricity is not pulsing, the states of the switches are what is changing. The electricity is always flowing through the circuit and used to change the states. Voltage and frequency are not related to each other in that sense. Increasing voltage increases power draw as in all electronics, but increasing frequency does as well. They are independent variables in this situation, we relate voltage so closely to frequency because voltage determines what is the real world maintainable frequency.

I tend agree more with the simplification. The electricity is pulsing - current draw is pulsing over time and as an unfortunate side effect, so is the voltage supply. Static CMOS only requires current while switching and a transistor switches in a fraction of the cycle time of the clock. As you increase the clock frequency, these switching events happen more often which creates more frequent power pulses. If you graph current versus time on a CPU's supply, it does look it's pulsing. Actually if you just look at the voltage supply over time, it's actually pulsing too because the voltage on the CPU's power rails droops.

 

[Diogenes2]

.....................


That is such an over simplification that it borders on incorrect. The electricity is not pulsing, the states of the switches are what is changing...........

When switches change state, current is being switched on and off . = pulsing

By any reasonable definition of the word ..

 

[betasub]

What you are doing with overclocking is trying to increase the frequency. To achieve that you must increase the voltage, the cost is always power consumption and heat.

Sorry for picking out a single point from your lengthy and helpful post, but increasing the voltage isn't always necessary. Indeed, some ATers have overclocked CPUs running with reduced voltage (undervolting).

 

[Daedalus685]

I tend agree more with the simplification. The electricity is pulsing - current draw is pulsing over time and as an unfortunate side effect, so is the voltage supply. Static CMOS only requires current while switching and a transistor switches in a fraction of the cycle time of the clock. As you increase the clock frequency, these switching events happen more often which creates more frequent power pulses. If you graph current versus time on a CPU's supply, it does look it's pulsing. Actually if you just look at the voltage supply over time, it's actually pulsing too because the voltage on the CPU's power rails droops.

What I read from the pulsing statement was that the current flow (thus the supplied voltage) is actually pulsed in terms of 'on and off' with the frequency.

I suppose it is semantical, I would reserve the term pulsing for behaviour such as clear cut on and off. The electricity is still flowing, regardless of the clock frequency, at all times. It is the power draw (and thus fluctuations in voltage) that go up and down with the transitions in state.

The actual act of transitioning the state is what consumes power, in a perfect world the draw would be zero otherwise (minor draw to read a state). So sure, there are more actively power consuming states per second with a higher frequency, regardless of the voltage setting. Thus frequency is roughly linearly proportional to the power consumption. But we don't increase the voltage to drive the frequency, we increase it to decrease the transition time of the circuit. Obviously that increases the per cycle power, as well as the power consumption associated with straight forwardly increasing the current throughout.

I should clarify that frequency and voltage are not entirely independent from each other in a physical sense, as increasing voltage increases power partly because of the transitioning. But they are individually controlled and affect different aspects of the system.

 

[Daedalus685]

When switches change state, current is being switched on and off . = pulsing

By any reasonable definition of the word ..

There is always a voltage supplied to the gates, though the power draw increases (in theory infinitely) during transitioning.

The digital signal will surely be a pulsing signal, up and down depending on the value.. but I was referring to the supply voltage, which is also only required during the transition (and when reading the data int the case of memory). Whether it is opening or closing it still consumes roughly the same power.

The electricity supplied is not a pulsed signal.. which is what my first reaction to his post was. I understand the semantics involved in this.. and can certainly respect the use of the term pulsing when looking at the current draw.. which I had not considered he may have meant. I've been working with pulsing power supplies lately.. I suppose I jumped to the conclusion that something analogous to this was what was meant, which is not likely the case.

 

[Daedalus685]

Sorry for picking out a single point from your lengthy and helpful post, but increasing the voltage isn't always necessary. Indeed, some ATers have overclocked CPUs running with reduced voltage (undervolting).

must was a silly choice on my part.

There are always tolerances in the voltage to ensure nothing can break. Often these tolerances are wide enough to clock something much higher on 'stock' voltage. Obviously 'must increase voltage' only applies in the situation where the maximum frequency for a given transition time is reached.

 

[Voo]

Sorry for picking out a single point from your lengthy and helpful post, but increasing the voltage isn't always necessary. Indeed, some ATers have overclocked CPUs running with reduced voltage (undervolting).

That's because "we need to increase the voltage to up the frequency" is a simplification.

A little bit more exactly (although still vastly simplified):
The whole problem is that we're working at the analogue level and our transistors aren't optimal. So when we "switch" the input from 1 to 0 (or vice versa) neither the output nor the input change immediatly.

I think a little picture should make that a little bit clearer: http://img23.imageshack.us/img23/2752/cmos.png

Well if the frequency is too high we can't distinguish zeros from one any longer and we have a problem.


@power consumption: I think we should just differentiate between static and dynamic power consumption and everything is fine
static power consumption: transistors are not ideal and do not switch off completly, resulting in a small leakage current that flows even in the off state
dynamic power consumption: gates and wires have capacities that must be charged or discharged during the switching process

Traditionally dynamic power dominated static power, but in the last years they contribute almost equal parts though static power c. is growing faster. So

The formulas for the interested:
f.. frequency
C.. wire capacity

Pdyn: 1/2 * C Vdd^2 * f
Psta: Idd * Vdd

PS: Please correct me if I'm wrong, most of that stuff is hidden in the darkest corners of my brain.. brrr I don't think anyone likes that analogue stuff

 

[EarthwormJim]

Electricity flows every time the cpu pulses. Pulsing more times means more electricity. Increasing voltage generates more heat for each pulse.

If your duty cycle remains constant, that's not necessarily true. Where the increase in power comes from is switching losses.

 

[Borealis7]

the CPU has a clock generator. each "tick" of the clock a "step" is executed. a cpu function is made out of several steps. and so, increasing the frequency of "ticks" gives more "steps" per second.

but, you must consider not all of the cpu's components are synchronized. some take more than one "tick" to finish their jobs (due to electircal latency or flip-flop latency) and if these components cant keep up with the clock you'll get calculation errors (for instance, the "answer" from register A was taken before the ALU populated the register with the real answer) and that causes data corruption, BSODs and crashes.

thankfully, talented electrical engineers work hard to get all these things to work together and allow you to grow your e-peen and brag about it.