Benifits of overclocking?

[kevinalm]

>>The reason it gets hot is because the physical material of the traces and the silicate substrate are resisting electron flow. It's literally frying the chip.

Partly true. In order to overclock it is usually necessary to raise cpu supply voltages above specs, which will directly raise power disapation. What is not commonly known is that the main cause of the added heat is simply the increased clock rate. Digital circuits heat during the transistion period between 1 and 0. (Digital circuits disapate little at logic high or logic low. It's in between that heating occurs.) Raising the clock increases the number switches per second and thus total device disapation. A minor technical point I always thought interesting.

[eats only heads]

here are some articals about over clocking.

http://arstechnica.com/tweak/oc_cooling.html

[Shadow]

http://www.TomsHardware.com

[Sayeh]

Partly true. In order to overclock it is usually necessary to raise cpu supply voltages above specs, which will directly raise power disapation. What is not commonly known is that the main cause of the added heat is simply the increased clock rate. Digital circuits heat during the transistion period between 1 and 0. (Digital circuits disapate little at logic high or logic low. It's in between that heating occurs.) Raising the clock increases the number switches per second and thus total device disapation. A minor technical point I always thought interesting.

No. Allow me to explain the rest of the _real_ details.

Actually, what you are failing to understand is that in order to raise the cycle rate, it requires higher voltage. The simple reason is that as cycle rate increases, it becomes more difficult to digitally force a waveform to change from a positive voltage (in this case 2.5v or perhaps 3.7v or even 5v) to neutral (or negative in some cases) and back again. It's a square wave.

In electricity, electrons only travel on the surface of a conductor. They do not travel through the center. Thus, as you raise the voltage, the diameter of the conductor (a trace) becomes insufficient to handle the current capacity and resistance causes heat to be generated.

Dissipation is an effect that occurs after the heat has already been generated. Keeping a chip cool doesn't negate the loss or eliminate the resistance. It merely hides the symptoms and prevents a physical materials breakdown.

[kevinalm]

>>Actually, what you are failing to understand is that in order to raise the cycle rate, it requires higher voltage. The simple reason is that as cycle rate increases, it becomes more difficult to digitally force a waveform to change from a positive voltage (in this case 2.5v or perhaps 3.7v or even 5v) to neutral (or negative in some cases) and back again. It's a square wave.



Sorry, but what you are missing is that I'm talking rise times and duty cycles. And it's NOT a square wave. It's quite true that the necessary voltage boost increases dissapation. (As I stated before.) But the major effect is subtler than that.

A transistor in the on state heats very little. (power dissipates in load) A transistor in the off state heats very little. (no current, no power to dissapate). It's in the intermediate state that a transistor heats. Since the switching time is more or less constant in logic circuits, heating is strongly dependant on frequency. At very high frequencies things get even worst due to the transistors never fully reaching cutoff or saturation, which creates "sneak" paths for the current to ground without ever leaving the chip to the load.



>>In electricity, electrons only travel on the surface of a conductor.

Just plain wrong. What happens is that when you raise the frequency from 0 hz up the current tends to be constrained to a layer on the surface of the conductor. Dc travels through the entire body of the conductor. After about 30 Mhz skin effect has done it's worst. Once you get into the 100Mhz region things don't change that much.


What's really important though is that when you overclock, things get hot very quickly.

[Sayeh]

Sorry, but what you are missing is that I'm talking rise times and duty cycles. And it's NOT a square wave.

Okay, okay-- let's start this over. First let me say that I am an EE. I helped develop Intel's 4004 CPU. I've worked on many others. Digital electronics is not a mystery to me at all.

I have tried to provide a slightly simplified view of what's going on because most of the details you've mentioned are meaningless to the "concept" of why things get hot, to most people.

In the first place, Direct Current (that's DC to you) is run through the processor at a specific, fluctuating voltage. It fluctuates from let's say 0v to 3.5v and back again, endlessly. The number of times this fluctuation occurs per second is the Processor Frequency. Electrically speaking, this is a Hertz rating.

In a 1KHz CPU, it would change the voltage state from 0v to 3.5v and back to 0.v again ideally 1000 times per second.

The reason the _SQUARE_WAVE_ waveform is important to understand here, is because it requires current to change the voltage state from 0v to 3.5v, or vice-versa. In digital electronics, this change must be nearly instantaneous. As such, on a scope, a square-wave is formed. AND IT IS IN THAT CHANGE where the "loss" is generated which becomes heat.

Here's why--

The more times you require that nearly instantaneous state-change, the more current it requires to make that change.

A CPU is built with certain expectations-- namely, they don't expect the chip to have more than a certain amount of current run through it. That maximum amount of current determines the physical characteristics of the components in the chip.

In electronics, if you run excessive current through a component, that excess current becomes a loss. In digital circuitry all such losses result in heat. This is a physical materials/characteristics issue.

You talking about "rise-times" and "duty-cycles" is the "how", not the what. I'm explaining the physical what.

[thedumbmutt]

lol, your posts are amusing when you get mad at someone, you just call them stupid and tell them to go away, anyone remember the president clinton thread?

[kevinalm]

Sayeh, re your last post. If you look back over the thread, you'll see that's exactly what I was saying. I was using simplified language for a more general audiance. I perhaps could have been clearer.

I still maintain there is no such thing as a square wave. Fourier kind of rules that out. I'm an E tech myself.

[eats only heads]

one thing to remember, Many chips you buy are just under clocked versions of faster chips. I heard for example that 4 out of 5 300 mhtz cpu's could be run at 450 mhtzs no problem. This is because intel underclocked them for the sake of the maket. If they just sold 450 mhtzs chips then the prices would drop too fast. None the less you can't run a 450 mhtzs any higher then 450 mhtzs.

[thedumbmutt]

that would ssem to explane why intel can oc without heating up as bad as an amd.

[Sayeh]

I still maintain there is no such thing as a square wave.

True, and I don't really disagree with that (after all, it is an analog world), it's just that we get as close as we can to a square wave. It's sort of like this analogy:

If you are standing 5 feet from Marilyn Monroe, and you keep cutting the distance between the both of you in half, at what point are you close enough for all intents and purposes?

cool.