OC Theory Question for Gurus

[ixtapalapaquetl]

I was wondering if anyone - maybe a EE? - could explain to me some of the theory behind overclocking a CPU.

 

Here's what I was thinking: any given chip runs comfortably at stock frequency and voltage. The frequency can be increased a certain degree, but at some point hits a wall because of lack of voltage. Why? That is, how does more voltage allow for higher error-free frequencies?

 

Then this cycle gets repeated; more volts, more hertz, more volts, more hertz... until another wall is hit ultimately controlled by temperature. But if temperature can be brought down (better fan, water cooling, liquid nitrogen, the depths of outer space...), the voltage-frequency cycle can run a few more iterations. But why does high temperature cause errors in calculations?

 

According to what I have described above, it seems that there might not be a hard ceiling for any given chips speed. Could one, in theory, maintain a chip at absolute zero, then crank it infinitely high? That can't be right though. So what are the physical limitations that dictate the ultimate maximum that one might reach?

 

I would love to know the answer to any of the above questions, and whether or not the model I described is accurate. Thanks. Nerds rule.

[century child]

Here is one of the best descriptions I have read on the subject. It is somewhat old but the principles remain.

 

http://www.gamepc.com/labs/view_content.as...enceofoc&page=1

[Sharp]

Hello,

 

Power per cycle.

The power is spread evenly across each cycle.

 

If you increase the clock, you generate more cycles per second.

Power per cycle is reduced, so you need more voltage(power) to compensate.

 

If you dont, the power per cycle will be to low and the wave(signal) may be to weak or not peaking at the correct voltage level.

 

+ everything else in the CPU is going to be running on a higher clock so you need more power.

 

And more power = more heat.

[JCamaro]

Here is one of the best descriptions I have read on the subject. It is somewhat old but the principles remain.

 

http://www.gamepc.com/labs/view_content.as...enceofoc&page=1

 

Cheers, extremely simply yet concise, good link

[evogre3n]

Then why do you reach a wall when you use coolling like dry ice or something, doesnt that lower the core temp to like sub zero?

[Fight Game]

like anything, input power is directly related to output power. cpu's are only meant to operate between certain voltages (and wattages, and amps, and impedence-ohms), and temperatures. You can reach a wall by either voltage or heat.

Interesting read up there.

[Dark Horse]

I think the easiest way to think of it is like this:

 

Simple Definitions:

 

Current: The actual amount or volume of electricity flowing

 

Voltage: The force pushing that electricty along.

 

When electricity flows through a semiconductor like silicon it encounters resistance as the current (or flow of electrons are pushed through it). The higher the frequency of the cpu or more "work" the processor does, the hotter the silicon gets.

 

As silicon heats up it becomes a worse conductor of electricity because the electrons flowing through it are disrupted by the heat energy in the silicon.

 

To counter this "disruption" or interference which is what causes errors the voltage needs to be increased to force the electicity through.

[evogre3n]

So, if I understand correctly,

 

overclock = more heat

 

more heat = more resistance so less current (amps)

 

apply more voltage = stabilize the current to its original

 

correct?

[Fight Game]

So, if I understand correctly,

 

overclock = more heat

 

more heat = more resistance so less current (amps)

 

apply more voltage = stabilize the current to its original

 

correct?

 

overclock = more heat, well depends on how high you oc! , but in general yea

 

I agree..

[Guest Dangermouse]

Great question! I just got a small education on transistors the other day, so I'll share what I can.

 

So think of a transistor like a draw bridge. The road going up to it and away from it is your electrical signal path. The bridge is the silicon. Silicon is a material called a "di-electric", meaning it can be a conductor or an insulator, depending on current. So when the silicon is insulating (bridge is up), electrons cannot pass to the other side. Flip a switch and the silicon becomes a conductor (bridge is down) and electrons can flow. So the problems you run into are:

 

more heat = less difference between conductive silicon (bridge up) and insulating silicon (bridge down)

more voltage = bigger difference between electrons flowing (bridge down) and electrons waiting (bridge up)

 

You should reach a wall eventually with voltage. Basically at modern production the "bridge" is only about a dozen atoms wide or so, so every once in a while an electron will shoot through to the other side, even when the gate is off (bridge is up). This happens more often as you increase voltage. So manufacturers (AMD and Intel both), go through the following stages: Increase voltage, increase speed, shrink transistor size, increase voltage, increase speed, shrink transistor size, etc. etc. We're basically at a point now where those three elements are as far along as they can get, so you won't likely see 4 or 5ghz processors next year. That's where dual-core comes in

[Gunkon]

Isolating the "heat and power" topics we reach others "Walls"... like Frequency response of the chip, Radio Interference, Frequency limits and others.

 

The frequency response of the Chip its read as a group of caracteristic curves wich reflect the capabilities of the chip Vs frequency, this factor is very important. you cant run a frequency beyond this limitis without errors and other ways of malfunctions.

 

The radio interference (and others) generated by a high freq clock may cause adversial effects on the surronding electronic elements.... this can be fixed with some techniques but limited.

 

And Isolating this factors....

 

Frequency limits.....

 

Almost all technology in computer electronics is based in conductor-semiconductor and aislant materials... they work with electrical current.

When you increase the clock freq more and more and more.... the wave length of the signal decrease and decrease and decrease.... and this electric signal its tranformed in light... at this point a copper wire is useless... we need Optical fiber... AND OUR AMD64 WILL BE UPGRADED TO WARP TECH OF STAR TREK....

 

P.D. (Sorry, My english is poor (Spanish native))

[evogre3n]

Great question! I just got a small education on transistors the other day, so I'll share what I can.

 

So think of a transistor like a draw bridge. The road going up to it and away from it is your electrical signal path. The bridge is the silicon. Silicon is a material called a "di-electric", meaning it can be a conductor or an insulator, depending on current. So when the silicon is insulating (bridge is up), electrons cannot pass to the other side. Flip a switch and the silicon becomes a conductor (bridge is down) and electrons can flow. So the problems you run into are:

 

more heat = less difference between conductive silicon (bridge up) and insulating silicon (bridge down)

more voltage = bigger difference between electrons flowing (bridge down) and electrons waiting (bridge up)

 

You should reach a wall eventually with voltage. Basically at modern production the "bridge" is only about a dozen atoms wide or so, so every once in a while an electron will shoot through to the other side, even when the gate is off (bridge is up). This happens more often as you increase voltage. So manufacturers (AMD and Intel both), go through the following stages: Increase voltage, increase speed, shrink transistor size, increase voltage, increase speed, shrink transistor size, etc. etc. We're basically at a point now where those three elements are as far along as they can get, so you won't likely see 4 or 5ghz processors next year. That's where dual-core comes in

 

awww, no 4-5ghz cpus!! darn