Timepass said:
problem is some of that stuff is impossible to do.
Namely the processor speed and the DVD drive.
The max speed a Processor can ever really go is around 7-8ghz since you are getting to around the max speed of an eletron around then.
As for the DVD they would spin to fast for the DVD to structully handle and they would fly apart.
It's possible to increase optical read and write speeds by using more lasers, working in parallel, instead of just increasing rotation speed.
portent said:
Not true at all. IBM demonstrated silicon-based transistors operating at 350 GHz, and the University of Illinois has demonstrated transistors at over 600 GHz.
These transistors can't be used to build entire processors yet, but there's no theoretical reason that we couldn't have a 300GHz processor one day. It may never be practical, but it should be possible.
It has nothing to do with the "speed of an electron." You don't speed up the electrons; you just move the components closer together. That way, the electrons have less distance to travel, and get there faster. The limit is in how small you can make the components.
Historically, most CPUs have been designed with synchronous logic, which means that all parts of the CPU have to work together in lock-step. The clock sends a pulse, throughout the CPU, that keeps everything in sync. Some CPU operations take a while to happen, and some happen quickly. But, the clock is constrained by the slowest operation. So, if it takes 1 nanosecond for the slowest operation to happen, then the duration between clock ticks must be at least 1 nanosecond, which means that the frequency of the clock is at most 1 gigahertz (1 billionth of a second -> 1 billion hertz).
So, what makes these operations take up so much time?
- Well, the CPU has to store state, in things like registers and control flags. Think of it like each storage location, for a bit (a 1 or 0), has to fill up with electrons, or empty out its electrons, which takes time.
- Also, electrons have to propagate across the CPU, from one part to another, along wires. The speed of light is around 300,000,000 meters per second, which is about 3e8 m/s in scientific notation. But, propagation of electrons in a metal is slower than light in a vacuum, so let's just say around 1e8 m/s. Now, a CPU is around 1 cm across, and the time it would take an electron to move from one side of the CPU, say where the clock is, to the other, where some state has to be stored, is about 1e10 cm/s. So, it would take 1/10 of a nanosecond, and thus the CPU could not run faster than 10 GHz.
Ok, so even if a transistor can change between open and closed states at 300 GHz, that doesn't change the propagation limit.
But, it's more complicated than that. First off, the clock itself, being an electrical signal, takes time to propagate. So, if you know every state the CPU can be in, and can calculate how the clock will propogate, then you don't need everything to be done before starting the next pulse, but only need it to be done before the next pulse hits that spot.
A much simpler thing to do is to give different parts of the CPU their own clocks, and make them run as fast as each part can go, and just make them talk when they need to pass info back and forth.
Which is really just a limited special case of asynchronous logic. With asynchronous logic, a part of the CPU doing work just has to signal when it's done, and not rely on a clock at all to keep things working together. Think of it like with the old synchronous logic, we plan before-hand to meet with our friends at 1pm and 3pm and 5pm. But asynchronous logic means when we're ready, we just call whoever needs to know something, with our cell phones.
Ok, so say we use the spiffy transistors that change state faster, and we make things smaller, so the propagation delays are small, then we still have current leaking, because with things so small, we're squeezing less electrons through smaller wires.
These thing might be solved by going optical, but that raises its own set of issues.
Personally, while I do think we'll get CPUs into the tens of gigahertz, I don't think it will go further, which is why most chip manufacturers have given up on that end, and are all going for multiple cores working in parallel.
