What determines the “computing power” of CPU cores? Why is an ARM core generally considered less “powerful” than an x86 core?

Question

We hear all the time that a smartphone with 8 cores is nowhere near as powerful as a desktop PC with 8 cores, and this is mostly chalked up to the smartphone having 8 ARM cores which are weaker computationally than 8 x86 cores, but what aspect of a CPU core design makes it better or worse at computing than another design? Hypothetically, if I wanted to make an ARM core as powerful as an x86 core, what would I change about it?

In: Technology

Answer 1

Okay, let me try.

First of all, for simplicity, remove the idea of “core” from the equation. A core is just a processor. An 8 core CPU is 8 processors stacked together. This is done because we’ve bumped into the limit of how good we can make a single processor, so we’ve decided “screw it, if we can’t make one 8 times better, let’s just use 8 of them”.

With that of the way, what is a CPU clock frequency? It’s a beat. Think of a guy banging on a drum to set a rhythm. That’s all it is. When the drum goes “bong!” something happens. But that beat is just there to keep things running smoothly. Nothing dictates how much stuff much happen from one beat to the next, so you can’t usefully compare a 2 GHz ARM CPU to a 2 GHz Intel one, or even Intel CPUs of different generations.

Now, what makes one CPU better than another? Loads of internal design stuff. Let’s say that the first CPU design we had couldn’t multiply at all. So if you wanted to calculate “2 * 500”, what you’d have to do is to “add 2, 500 times”. This obviously takes at least 500 times as long as a single addition. So that’s slow, inefficient. For the next CPU we figure out how to create a multiplication instruction, and it takes say, 4 times as long as adding (because multiplication is a more complex operation). So now “2 + 2” takes some amount of time, “2 * 2” takes 4 times as long, but “2 * 500” also takes 4 times as long. We’re improving! And in the generation after that, we designed some really fancy circuitry to make multiplication happen exactly as fast as addition.

Repeat this kind of thing, many times, and we’ve got the hell of an improvement.