Why do we need to increase voltage when overclocking PC components?

[Delta2]

TL;DR Summary

Why overclocking usually also needs to overvoltage?

When we overclock CPU or GPU or RAM we also have to increase the voltage to achieve stable operation (provided we have adequate cooling). Why is that?

 

[Tom.G]

Because during switching, it takes time for the signals to reach their final voltage.

If you tell them to switch more often, they eventually don't reach their needed 'threshold voltage' before they are told to switch back again. The 'threshold voltage' is an internal design parameter in the chips and is the 'decision point' where the chip decides if the signal is High or Low. It is relatively stable over supply voltage.

Just like a DC motor, the higher the applied voltage, the faster the signals will change level; and the sooner the chips detect the level change.

That's the 'easy' version, the full version needs a lengthy study in physics, electronic design, and solid state physics.

Hope it helps!

Cheers,
Tom

 

[Keith_McClary]

A kid was shocked to learn that there was a clock to slow down our computer! Why not just let it run infinitely fast?

 

[berkeman]

A kid was shocked to learn that there was a clock to slow down our computer! Why not just let it run infinitely fast?

Was that kid you? Otherwise, how does your post help this thread?

 

[Keith_McClary]

how does your post help this thread?

It's another aspect of the same limitations of the electronics.

 

[berkeman]

Wakarimasen. I don't understand.

A kid was shocked to learn that there was a clock to slow down our computer! Why not just let it run infinitely fast?

Please explain your reply in detail with technical details and reference links. Thanks.

 

[Baluncore]

There is a class of logic gates that do not need a clock. They employ more than one conductor per bit, and different combinations of signals have either a valid or an invalid state.
The output becomes invalid when any input changes. Once the output combination becomes valid, it is correct. See; Asynchronous Logic.
https://en.wikipedia.org/wiki/Asynchronous_circuit#Asynchronous_CPU

 

[Delta2]

Sorry guys but I find posts #3 to #7 off topic. @Keith_McClary is probably trolling the overclocking community but I don't understand why @Baluncore decided to throw in the case of asynchronous CPU. OK no clock there so not possible overclocking at least not in the traditional sense.

 

[Keith_McClary]

Please explain your reply

The answer to both questions is that the circuit requires a certain minimum time (which may depend on voltage) to complete a cycle.

 

[Delta2]

The answer to both questions is that the circuit requires a certain minimum time (which may depend on voltage) to complete a cycle.

How this minimum time to complete a cycle can depend on voltage?

 

[Keith_McClary]

How this minimum time to complete a cycle can depend on voltage?

As explained by Tom.G

 

[Baluncore]

How this minimum time to complete a cycle can depend on voltage?

You can run a CPU at the specified clock rate reliably. You may be able to run it faster, but at some speed, voltage, and temperature, it will become unreliable.
CMOS logic swings between the supply rails.
1% higher supply voltage offers 1% higher gate voltages.
Gate voltage 1%, determines CMOS channel conductance 2%, (square law).
Gate outputs slew through 1% greater voltage, with 2% more conductance, so is only 1% faster.
But the 1% greater voltage and 2% greater conductance makes for 3% greater power.
So the overclocked chip only goes 1% faster, but it produces 3% more heat.
Thresholds drift with temperature. That is not a recipe for reliability.
You have no idea how often an overclocked CPU is doing a bad computation.

but I don't understand why @Baluncore decided to throw in the case of asynchronous CPU. OK no clock there so not possible overclocking at least not in the traditional sense.

While the temperature and voltage may change, an asynchronous CPU always runs reliably at the maximum possible speed.
You know that an asynchronous CPU is going as fast as is possible, and that it is not doing bad computations.

 

[alan123hk]

When we overclock CPU or GPU or RAM we also have to increase the voltage to achieve stable operation (provided we have adequate cooling). Why is that?

I think that an in-depth study of the actual calculation formula of the propagation delay time of CMOS circuits will help us understand why increasing the power supply voltage can speed up the switching speed.

Please refer to the following link:
http://web.mit.edu/6.012/www/SP07-L13.pdf

Obviously, when the power supply voltage increases, according to the relevant equations, the propagation delay from high to low and the propagation delay from low to high will decrease respectively.

Unfortunately, the power loss will also greatly increase in proportion to the square of the power supply voltage.

 

[Baluncore]

Unfortunately, the power loss will also greatly increase in proportion to the square of the power supply voltage.

The power equation on page 13 shows that a 1% increase in Vdd increases power by 2%, (square), and the 1% increase in frequency makes it a 3% increase in power.

 

[alan123hk]

The power equation on page 13 shows that a 1% increase in Vdd increases power by 2%, (square), and the 1% increase in frequency makes it a 3% increase in power.

I totally agree with you.

That's it, because the power loss is proportional to the square of the power supply voltage, so 1.01² = 1.02, namely 1% to 2%.

If we assume the gate threshold voltage (V_tp) is much smaller than the supply voltage (V_dd), then the propagation delay reduction is approximately equal to 1.01/1.01² = 0.99, which means that 1% increase of voltage can increase 1% of speed or frequency accordingly, so the total power is increased by 1.02*1.01 = 1.03 (3%)

By the way, I think it is possible to estimate how much voltage we should increase under a certain level of overclocking according to the calculation formula of the propagation delay time.