How computer processors (and computers) get faster nowadays

724 views
0

Some 15-20 years ago it was all about CPU frequency – higher frequency meant faster computer (simplified). Then it was all about cores – more cores in CPU meant faster computer (simplified again). For quite some time now the CPU frequency is (based on physical limitations) not rising (it stays at +/- 2-3GHz) and number of cores in regular computers (read desktops or laptops) is the same as well (2-8 cores). But still, every year new and faster microprocessors are made, every year regular computers are getting faster. How so?

In: Technology

Well nowadays you are trying to shrink the “interruption gate” which basically means if that command comes as a 1 or a 0, CPUs are binary so it either let’s the electron through or not causing a true or false statement and you get millions and even trillions of that per clock, or IPC.
The more you can push through in a single clock the faster the calcutions are. We are pushing for smaller and smaller gates, we had 22 nm not too long ago and now we are pushing towards 10 nm and even smaller.
The smaller the gate is the faster and more gates that you can push into a same sized core giving you basically faster and effectively “stronger” CPUs.

Sorry for probably not using the correct names. I’m tired and kinda forgot everything xD

Cores are still going up. You can have 16 to 32 in consumer grade chips these days.

Also, they are becoming more efficient. More cores does not always mean more power. That is why we have single core workload tests. One intel core is usually more powerful than one AMD core because of how efficient they are. Though AMD has recently had little bit of a comeback in this regard.

The processors can be made faster per cycle (what the frequency measures), and generally gain a few percentage points of performance per generation at the same clock speed. One thing allowing this is the shrinking of transistors, allowing more to be used, but general innovation also occurs. The newest line of Intel processors, for example, is supposedly more than 10% faster per cycle on average, at least according to Intel’s benchmarking. There are also technologies like SSE and AVX that get added on occasion to speed specific tasks up. SSE/AVX specifically allows you to do operations on multiple numbers at the same time, so instead of adding two numbers each cycle, you can add 8 numbers to 8 numbers.

Both core count and frequency (and importantly frequency per core) have been generally going up, although slowly, which has been speeding up recently thanks in part to the competition between AMD and Intel.

There are a few factors that affect the speed of CPUs.

1. Temperature – higher temperature physically makes your CPU slow down. The more you can cool off your CPU, the more optimally it will perform.

2. Number of cores – more cores means your CPU can do more things at the same time, but code has to be written to take advantage of this, and most apps just don’t need multiple cores.

3. Transistor size – a core is made of transistors, and the smaller you make them, the less heat your CPU will emit for a given amount of work. But the smaller you make them, the more error-prone your CPU becomes due to quantum physical effects

4. Circuit design – typically a CPU is a CPU, and they are interchangeable to some degree. But this is not required, and you can make specialized CPUs, such as those for graphics processing, motion processing, security, etc etc. The closer a circuit represents the work being done, the more efficient it will be at performing that specialized task.

5. Clock speed – increasing clock speed gives you a temporary speed boost, but increasing the speed will inevitably increase the temperature which will decrease the speed unless you have a really nice cooling system.

Moore’s “law” states that the size of a transistor is cut in half every two years, and it has been correct so far. What this means is CPUs are running less hot, and therefore more optimally.

But eventually we will hit a limit on transistor size (or discover new physics) and the pace of faster CPUs will come to a grinding halt.

New physics example – If we can make a superconductor CPU at room temperature, we can eliminate the need for cooling because superconductors do not emit heat. The catch 22 is that (most) super conductors don’t exist until temperatures reach absolute zero (the coldest possible temperature) and supercooling is not cheap nor portable.

Edit: grammar