Why are the GHz speeds in computers the same after such a long period of time?
In: Engineering
We’re reaching size limitations. Our computers are so fast that the speed of electricity (a decent portion of the speed if light) itself is hindering them. Thus, we need smaller systems, but smaller systems are encountering issues with being unable to prevent electrical flow due to quantum effects.
For something to oscillate past a few GHz, the length of time between two signals has to be short. That time is now *not long enough* to cross the entire length of the silicon chip. This means that one signal can “overtake” the one in front of it, which causes absolute confusion and merry hell as we design chips to be “synchronous” (i.e. things happen at the same time everywhere).
We don’t have asynchronous chips (we can have multiple chips that aren’t in sync, but one chip tends to have a single concept of “a clock signal” that turns on and off regularly and makes everything else happen).
Past about 5GHz, the length of the pulses needed for that clock signal mean that, even at the speed of light, they can’t make it across the physical length of the silicon chip before another one starts its journey.
Making the chips asynchronous, shorter, or quicker actually makes things incredibly complex and liable to all kinds of problems if there’s a bug found later on. Not to mention, the higher the clock speed, the more heat given out (because the power required to make more oscillations is greater), which means more cooling or more problems with heat, and more interference.
Pretty much, we’ve hit a physical boundary that you can only compensate for by making chips tinier and tinier (which has other problems, not least manufacturing), colder and colder (supercomputers are sometimes liquid-helium cooled or similar), or more and more complex to design, produce, run, program and diagnose.
We’re hitting the upper speed limit for processors because of limitations like the speed of light, and we are having trouble making things smaller because on that scale quantum mechanics starts doing unexpected or unwanted things.
We are overdue for a discovery that will revolutionize computer processing yet again.
So for the past decade the focus hasn’t been to increase speed, but to increase efficiency. Processors are being made with increasing numbers of cores so they can do more at once, and bus speeds are increasing so the processor can talk to devices and RAM more quickly. All of these things translate to improved performance.
Rather than increasing processor speed, which is becoming increasingly difficult thanks to things like substrate bleed (which is a whole other conversation), the push hasn’t been to increase clock speed (measured in Hz) but rather to simply add more processor cores. As software development has matured and proper utilization of multiple cores to get work done has become commonplace, the value of number of cores has steadily outpaced raw clock speed.
Clock speed used to be king because there was only a single “pipeline” at work in the processor. Stuff went in at one end, did what it had to do, and came out the other end. The faster you could get through the pipeline, the better. Modern processor architecture has added more and more pipelines running together. By spreading what’s coming in across multiple pipelines, it keeps everything flowing more smoothly than trying to stuff it all through one pipeline.
Additional factors include decreasing cost and size of what’s called *cache memory*. This is memory that’s actually on the processor itself and is used to store data the processor is actively using. It’s far, far faster than having to write data to system memory and retrieve it. Between increased cache memory and more effective use of multiple processor cores, the importance of raw clock speed has sharply dropped off over the past 10 years.
I haven’t seen it here yet for some reason, but one of the biggest reasons is heat from power consumption. Processors get unsustainably hot because they are less efficient as power consumption increases.
For decades, if people’s computer programs were too slow, they would wait a year for processor speeds to increase in order to get a “free lunch”.
Simply put we can’t make them any smaller. For the longest time & for most of the speed up, computers got faster because we could make their most basic part(a transistor) smaller. The smaller the parts, the smaller the distance electricity had to move to make things work.
But now we’ve gotten to the point that the transistor is only a few atoms, and any smaller & it just won’t work.
Lets say you own a deli shop making sandwiches. When you have a large order there are 2 ways to get the job done quicker. One is to make sandwiches faster (ghz). The other is to hire more people to make the sandwich (core). Current technology it is just cheaper to hire more people than to get people to work faster.
Processors can’t multitask. What they can do, however, is rapidly switch between different tasks. Pushing clock speed (Ghz) higher has diminishing returns for the amount of effort and cost involved, in getting better performance from the computer. Instead, its better to have multiple cores, each one capable of doing its own task so collectively, the CPU as a whole is multitasking. This means you can play a video game, for instance, having one core dedicated to that, and another to a background task so they aren’t competing to use the same core.
A lot of this has already been answered, but let me provide a bit of perspective from closer to the silicon level since I’m currently on an internship working with this issue. While clock speeds are important, they are not the only factor in computer performance. Thus, current designs aren’t focused solely on increasing clock speeds.
One of the main issues is simply heat. As we increase the rate transistors switch the power required increases exponentially and it gets difficult to cool.
A more fundamental issue is that transistors and associated wire have capacitances, or the ability to store electrical charge. This effectively slows down the rate you change your signal- as the electrons in these reservoirs counteracts any changes you make until the electrons it holds is depleted. This makes a nice sharp clock signal flatten out and slows down rise times.
Lastly, it is difficult to design good interconnects. Even if we have a really high clock speed, it’s not easy to design wires that can carry information at that clock speed. All wires have some capacitance and inductance where energy is temporarily stored in electric and magnetic fields instead of being sent down the wire. Worse still, the magnitude of this energy that is stored is frequency-dependent. This means at higher clock speeds/frequencies a lot more energy is “lost” before getting to the end. This means that the magnitude of the signal at the end is a lot less. For example, one thing you see is that at higher frequencies, signals on one wire start leaking to other wires close by- something you obviously want to avoid.
Let’s pretend your family is moving across town and you have 5000 boxes of toys. The moving truck can only fit 50 boxes at a time which means your dad will have to make 100 round trips. The fastest your dad can drive on the freeway is 65mph. Sure, he could drive at 100mph to make better time but damn it, he loves you too much to risk jail time. So instead of pushing himself to drive faster on the freeway, endangering himself and others, he decides it’s a better idea to have your mom drive a second moving truck, also packed with 50 boxes of your awesome toys. Now the two of them only have to make 50 round trips each, halving the initial time it would’ve taken, all without the risk of a speeding ticket or going to prison! Now imagine how faster the move would be if your parents also recruited your uncle Bob and aunt Sally to drive a third and a fourth moving truck. They would be able to move all of your toys 4 times faster than if your dad had to move everything by himself! To match this speed by himself, your dad would have to drive at 260mph and the U-haul down the street isn’t renting out Koenigseggs yet.
Latest Answers