2
\$\begingroup\$

When you increase the voltage into a computer's CPU it increases the clock speed of the processor and can lead to better performance. Why does the increased voltage cause the processor's clock speed to change?

\$\endgroup\$
  • \$\begingroup\$ "...it increases the clock speed of the processor..." [citation needed] \$\endgroup\$ – Ignacio Vazquez-Abrams Aug 18 '15 at 1:24
  • \$\begingroup\$ I think that it is the other way around. A higher clock speed needs a physically smaller IC, and a smaller IC needs a lower voltage. \$\endgroup\$ – Roger C. Aug 18 '15 at 1:30
  • \$\begingroup\$ You don't increase frequency when you increase power. But if you increase frequency, you have to increase power to make your CPU work. It require more "juice". But thermal dissipation get bigger, wich make it hotter, more ressitance, more heat... and so on. \$\endgroup\$ – bokan Oct 19 '15 at 13:22
9
\$\begingroup\$

A CPU is, at its root, an extremely large sequence of FETs driving each other in complex combinations. These FETs have a very large number of possible states, only some of which are valid. Each clock cycle, the processor should move from one valid state to the correct succeeding valid state. The FETs are arranged such that they begin to switch from one state to the next only on a clock edge. But if one FET drives another, that means the driving FET has to complete its transition before the load FET can even start!

This is propagation delay, and the longest chain of propagation delays in the system determines your maximum clock frequency. You have to be sure that by the time the next clock edge arrives, all FETs everywhere have completed whatever transitions are going to occur. Otherwise you end up with your chip in an invalid or incorrect state, and all bets for correct operation are off.

So the minimum clock period (and thus maximum frequency) is a function of the longest propagation path in the chip, and of how long the individual transistors in that path take to switch. You can't change the propagation path once the die is etched, but you can change the time it takes for a transistor to switch. As with any FET, the switching time is affected directly by the time it takes the gate capacitance to charge. Since the capacitance is fixed, a higher voltage through the same FET results in more current flow, and a faster rate of charge. So increasing the rail voltage of the processor can increase the switching speed.

Of course, the down-side is that at higher voltages, the switching loss of the FETs also goes up, more than the gains from faster switching times make it go down. So higher voltage results in increased operating temperature, which can also affect the switching time of the FETs, and ultimately result in damage to the chip.

Also, quantum wibbly-wobbly effects can make the chip behave unreliably.

\$\endgroup\$
  • 2
    \$\begingroup\$ Yup, speed increase is linear while power increase is quadratic. \$\endgroup\$ – Ignacio Vazquez-Abrams Aug 18 '15 at 2:46
  • \$\begingroup\$ "Since the capacitance is fixed, a higher voltage through the same FET results in more current flow, and a faster rate of charge". Yes, but the voltage in the "load capacitor" would also need to reach a higher threshold (to be considered a '1' for instance). Just like in a RC circuit where the constant time tau is independent of the charging voltage. Am I overlooking something? \$\endgroup\$ – Roger C. Aug 18 '15 at 13:02
  • \$\begingroup\$ @RogerC. I see you reasoning, and I don't have a definite answer. Best guess is that since the FETs are on "harder", their effective resistance goes down, reducing the time constant. Also, the threshold voltage of the FETs hasn't changed, so while it might take the same amount of time for them to reach their new ultimate voltage, it will take less time for them to reach the threshold voltage and start to switch. \$\endgroup\$ – Stephen Collings Aug 18 '15 at 14:08
  • \$\begingroup\$ @StephenCollings your explanation does make sense. Thanks! \$\endgroup\$ – Roger C. Aug 18 '15 at 14:19
0
\$\begingroup\$

Without going in to details when you design a chip you spend a lot of time doing timing analysis between all the little logical and signal paths to each cell inside the chip. You can only run a chip as fast as it will meet this timing without getting errors.

This max speed is also affected by: process, temperature, and voltage. So you can either by design or by trying to over clock increase the voltage to increase performance.

Raising the core voltage raises the voltage to each cell of the chip and this decreases the prop delay through that cell in effect making the paths faster. You can then increase the clock rate separately to see if your chip still functions error free at this speed.

Of course you can break timing this way too or break your chip by damaging it with over voltage or over temp.

Your process and temp will affect things too so you could get lucky and get a chip that can already run faster than it's clocked, or unlucky and get one already running at near max speed.

In the before time people would cool down chips too to get them to run faster but now at the smaller process nodes we have temperature inversion so they run faster hotter.

\$\endgroup\$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.