why does CPUs/CPU cores seem to be capped at 3 – 4 Ghz since almost a decade?
In: 1368
CPUs come with a base clock of between 3 – 4 GHz because it’s most efficient. As you increase clock speeds you must increase voltage as well. Part of the power equation is Voltage ^ 2, meaning that when you increase voltage a little, it impacts power draw more than increasing current or capacitance. More power draw equals hotter processors, but with Lithographic node shrinks you can lower the voltage needed to flip a transistor a little, freeing up more voltage to be used to push clocks higher, hence the “new node = more efficiency at the same clocks” or “new node = higher clocks at the same power level”.
TL;DR, it’s to cut down on power use. My Intel i7-11700k is built on 14nm+++ and uses 95W when clocked at 3.6GHz (it’s base clock) and draws 220W when clocked at 5.0GHz (boost clock that I’ve set to run indefinitely).
Not a ELI5. But aside from heat issue, digital design makes certain assumptions.
Two assumptions that I can think of start to break down at higher frequency. (1) Assumption about the wire transmitting the signals being short relative to the wavelength. As you go to higher frequency, the wavelength becomes shorter and this assumption starts to breaks down. (2) There are electromigration requirements. As you go higher frequency, the wires conduct more current on average and they break down faster. In theory you can make the wire wider to support the additional current but the logic gates that digital designers use usually favor being compact rather than supporting large current. Thus they cannot go too fast.
Chips or sub-components inside chips that go higher than 5GHz exist. But they are designed by people with different skill set. Usually they are called analog mixed signals (AMS) engineer or radio frequency (RF) engineer, depending on the speed and applications they are targeting.
Damn these answers suck. So there are a couple of factors here. Probably the biggest one is the death of [Dennard scaling](https://en.wikipedia.org/wiki/Dennard_scaling). Basically, shrinking transistors no longer provides the same speed/power benefits as they used to, and compounding the issue, the rate at which we see those shrinks has also slowed down.
But beyond that, in the lead up to the 3-4GHz processors we know today, there was a lot of focus on frequency. Specifically in the mid-late 90s through the early 2000s, clock speeds were increasing *faster* than Moore’s law or Dennard scaling would suggest. This was all from design and architectural work focusing on making faster circuits, and splitting the work between more and more (smaller) steps. But eventually, the “low hanging fruit” there started to run out, and coupled with the aforementioned process issues, we got significant stagnation in frequencies. That said, frequency scaling is not completely dead. We’re seeing chips regularly pass 5GHz today, and 6GHz should be doable out of the box within a year or two. Might even hit 7GHz by 2027-28.
There are, of course, other considerations than the ones I mentioned, like poor wire scaling, power density, etc., but they’re not the dominant factors.
Because electronics use what is called a square wave. Basically a signal that is a 1 or 0, on or off to make the logic happen. This has to have a sharp square edge. When a circuit starts getting around 5 GHz and higher it starts to round the edges of that square signal and at a certain point the digital signal starts to act as an analog signal which has round none strictly defined edges. When that happens all sorts of things go wrong because different parts of the circuit think the signal is on, off or not even there.
Until about 2006 every time transistors got smaller they also reduced in voltage and current substantially which together are power. This reduction in power allowed higher frequencies to “use” some of those power savings but for the last 15 years the power reduction hasn’t been as good as the area reduction. We can put more transistors in a given space but they use more power and so you see more transistors being used for more cores not higher frequencies.
[i9-13900k can boost up to like 6GHz or something](https://www.google.com/amp/s/wccftech.com/roundup/intel-core-i9-13900k/amp/) so we are still getting further and further every year on boost speeds
Thermodynamics and technological restraints only tangential to computing means heat can move only really so fast. Computer fans can only do so much, a liquid coolant can only pull heat out of your processor so fast, but mainly heat will only “voluntarily” move out of the computer chip(s) so fast.
But producing heat has essentially no such limitation.
Each computation creates a little bit of heat, via the resistance in the metal wires and resistors in the computer chips. The faster your clock speed, the faster you’re telling the computer to do computations. The clock speed being too fast for the computer to synchronize or whatever isn’t a problem nowadays, computer chips are small and the speed of light is pretty high at this scale, but the heat stays a problem.
Short answer: higher clock speed means the computer produces more heat, and our cooling technologies aren’t strong enough to pull heat away from the silicon to prevent them from literally melting as heat builds up. So we limit our computers so our cooling systems can keep up.
Picture this: you’ve got a pile of rocks, and you need to move them all to a spot 10m (or yards or 30 bananas or whatever) away. You have 1 guy sprinting back and forth as fast as possible, they’ve trained for this. Lots of energy but undeniably fast, he can go back and forth 5 times in 10 seconds. That’s pushing the cycles beyond 3-4GHz. Now imagine the same situation, but instead of 1 sprinty boi you’ve got 5 people jogging. Takes them 10 seconds to jog back and forth once. Still gets 5 rocks over the line in the same time though, except they spend a lot less energy overall, they can make do with a good sammij rather than high-end workout food.
Heat management.
Here’s an article about a potential future semiconductor that has way better heat transfer.
https://scitechdaily.com/mit-discovers-semiconductor-that-can-perform-far-better-than-silicon/
Latest Answers