Is there a speed limit for CPU’s? Could clock speeds exceed terahertz? How would this change current computing?
CPU’s use a “pipeline” which is several bits of hardware all in a line, each executing part of the instruction. The limiting factor is the speed of the slowest logic stage in the pipeline. If you want to make the CPU run faster, then you need to make the logic gates work faster. Making faster logic is an art that has these days almost progressed to complete voodoo, but basically you are fighting transistor switching speeds which are limited by things such as the physical size of the transistor and the inherent capacitance in its internal junctions.
When I went to college, 1 MHz CPU’s were state of the art. One of my professors was fairly certain that we’d hit physical limits that would prevent us from ever getting much faster than about 40 MHz. Now that we are approaching 4 GHz (aka 4,000 MHz), similar predictions about the “fastest” that a CPU will ever get always continue to amuse me.
Engineers are always running into limits, and always finding creative ways around them. Will we ever get into the terrahertz range? Seems almost impossible to me, but then again so did gigahertz once upon a time, so I’m keeping quiet.
Okay, but how about 20 or 30 Ghz?
There is a theoretical upper limit to the speed of CPUs, which is the speed of electrons (or holes) in a piece of semiconductor. That’s why GaAs chips are so much faster than silicon chips because of this. However, SiO2 being a natural insulator is a great advantage in the fabrication process, so much so that most mainstream chipmakers are sticking with silicon.
I’m with ECG, the next physical limit is probably the speed of light and I’ll bet that someone will come up with a different method of processing by the time we hit that.
Remember there are different ways to measure ‘speed’ in a CPU. The 1Ghz number is merely a measure of the computers clock rate. Think of it like a metronome. Each ‘click’ of the metronome causes everything in the CPU to move ahead 1 step. The more stepd you can squeeze into a given timespan the faster (generally) the chip is. There is ultimately an upper limit to how fast the clock rate can go. At the very least you have the Planck Time limit hanging out there which would absolutely put a halt to further speed increases but I’d wager the practical speed limit is somewhere far slower than that.
The other side to how fast you can make a CPU is how much ‘stuff’ gets done in a given step. Computers such as IBM’s ASCI White (the second fastest computer in existence) are massivley parallel. If you can’t speed up one chip then just get a lot of chips working together. In the case of Intel’s Pentium IV they increased the length of the pipeline (the true heart of a CPU) to allow greater clock speeds. When the pipe was shorter the rest of the chip couldn’t speed up as too much was being forced through the shorter pipe at a given moment. By lengthening the pipeline more things are in progress thus easing the pressure on the rest of the chip so higher clock speeds can be achieved. As with most things this comes with a tradeoff. CPUs use a thing called branch prediction to guess what the next operation should be before it is actually told what it should do. If it predicts correctly the CPUs speed is dramatically improved. If it gets it wrong the entire pipe has to be purged and start again. With the longer pipe you lose more when you have to dump the instructions in the pipe. For this and a few other reasons a higher clocked Pentium IV runs at about the same overall speed (i.e. ability to do work) as an AMD Athlon chip clocked several hundred Mhz slower. So why would Intel do such a thing? Unfortunately most consumers aren’t educated enough about this and they buy the chip with the biggest number attached to it. To fight this AMD has dumped rating its chips with a Mhz/Ghz moniker and has now taken to just giving them a number which happens to realte to what they feel is the chips equivalent performance if stacked against an Intel chip.
Let’s here it for marketing!