Understanding CPU Frequency Waveforms and Dead Time in Modern Processors

AI Thread Summary
The discussion centers on the challenges of generating square waves at GHz frequencies, particularly in modern CPUs. It highlights that at these frequencies, inductance causes waveforms to resemble sine waves rather than perfect squares. The conversation delves into the implications of clock frequencies on CPU performance, noting that increased frequencies can lead to crashes if previous instructions have not completed before new ones initiate, akin to "shoot through" conditions in power supplies. The role of gate capacitance is emphasized, as transitions in CPU states require charging or discharging capacitors, resulting in trapezoidal waveforms and increased power supply current, which can lead to overheating and faults. Additionally, as integrated circuit feature sizes shrink, leakage currents become a significant factor in power consumption, impacting performance and efficiency. The concept of a "sweet spot" for die size is mentioned, suggesting that while advancements continue, the optimal size may have already been reached or surpassed.
artis
Messages
1,479
Reaction score
976
There was a thread recently in another subforums where a user talked about creating "square waves" at Ghz frequencies. It was said there that at that frequency range one cannot have a square wave as the inductance "rounds off" the edges making the waveform sine like.
I am wondering then how about modern CPU's running at a few Ghz typically, what is the waveform like for them and how does the dI/dt look like?

what are the typical "dead time" between pulses as I read that increasing the clock frequency usually makes the CPU to crash due to the fact that the previous pulse hasn't settled yet "waveform dropped to zero", while the next pulse is already rising which disrupts the binary code as the states of transistors overlap, this sounds somewhat similar to a condition known as "shoot through" in switch driven power supplies (smps) where both/all switches conduct at the same time due to the overlapping of their conduction time.
 
Computer science news on Phys.org
A CPU running at a few GHz will have gate delays of about 10 picosec so there is plenty of time for the chains of gates to complete before the next clock phase. Most clocks inside a CPU are two phase clocks, not simple square waves with edge triggered registers.

Shoot through current is not the problem. The problem is that every transition must charge or discharge a capacitor. The gate capacitance of the CMOS inputs must be charged by the CMOS output current, so everything internal becomes a trapezoidal wave. The resulting charge flow is power supply current. That higher current heats the chip to the point where the safety margins are reduced and faults start to occur.

The reason why over-clocked CPUs crash is simply because a previous instruction has not completed before the state of the machine is changed for the next instruction.
 
  • Like
Likes berkeman and sysprog
artis said:
this sounds somewhat similar to a condition known as "shoot through" in switch driven power supplies (smps) where both/all switches conduct at the same time due to the overlapping of their conduction time.
In addition to Baluncore's reply, as IC feature sizes ("geometry") get smaller and smaller, the leakage current in each cell becomes a greater fraction of the power consumption. The shrinking gate sizes let you drive the gates faster, but there comes a time when the increasing leakage current becomes about the same as the clocking currents ##C \frac{dv}{dt}##
 
berkeman said:
The shrinking gate sizes let you drive the gates faster, but there comes a time when the increasing leakage current becomes about the same as the clocking currents
So there is a point where the "sweet spot" of nm die size is hit, I would say we are already there or maybe even a little past are we not?
 
artis said:
I would say we are already there or maybe even a little past are we not?
We have always been there. The "sweet spot" as you call it, moves with time.
 
  • Like
Likes pbuk, berkeman and anorlunda
In my discussions elsewhere, I've noticed a lot of disagreement regarding AI. A question that comes up is, "Is AI hype?" Unfortunately, when this question is asked, the one asking, as far as I can tell, may mean one of three things which can lead to lots of confusion. I'll list them out now for clarity. 1. Can AI do everything a human can do and how close are we to that? 2. Are corporations and governments using the promise of AI to gain more power for themselves? 3. Are AI and transhumans...
Thread 'ChatGPT Examples, Good and Bad'
I've been experimenting with ChatGPT. Some results are good, some very very bad. I think examples can help expose the properties of this AI. Maybe you can post some of your favorite examples and tell us what they reveal about the properties of this AI. (I had problems with copy/paste of text and formatting, so I'm posting my examples as screen shots. That is a promising start. :smile: But then I provided values V=1, R1=1, R2=2, R3=3 and asked for the value of I. At first, it said...
Back
Top