What's the limit of CPU evolution ?

[Mod]

GPUs are evoluting too fast. Its moore law is of 12 months ( period to double the number of chips ). The transistor miniaturization doesn't keep this pace, so NV30 GPU will have 2x the size of P4's . So, here's a couple of questions:

1)When the chip size will limit the evoulution of GPU moore law ? That is, the number of transistors increases more than the deacreasing caused by the transistor size. I don't think you can have a 1m^2 chip :eek:

2)The day when the scientists and engineers won't be able to improve the speed or performance of the chip by any physical means, what will happen to the computer industry in general ? ( use your futurology skills ) Consider that this physical means include quantum processing, and all weird things.

[Bigus Dickus]

To address #2... I'm not sure that it is even possible to have an end to physical improvement, not within the forseeable future. Even with today's CPU technology, it's only a matter of cost as to how much parallelization you can get.

In the distant future, quantum computing will change all that. Processing power goes up exponentially as each new logic block is added due to the parallel nature of the quantum world. In comparison, at best, current CPU power only scales linearly with additional logic blocks, under ideal circumstances.

[Harnagel]

1. As far as chip size is concerned I would guess we are pretty much at the limit with the NV30. You will notice that the nV30 goes at a much slower speed than a p4 (400 to 3000), but the trade off for die size to speed probably won't get much better.

2. Extremely hard to tell. I would say the limits of transistor gates will probably be around 1-10 nanometer as to the current 130. At moore's guess (double per 18 month last i read) it will take between 20-30 years to reach such dimensions. Lots of problems though concerning quantum effects and what not at these sizes. If we make some breakthroughs in nanotech then we might even start looking at 3d chips instead of 2d designs.

Then there is the whole quatum computer idea. If we get this working in multiple qubit arrays then there will be many advances in computational power. Esp for factoring algorythms ie unencrypting and such. Here the speed increases become exponensial to the qubits used. For example if a 128 qubit array was used to decypher a 128 bit encryption it would take 1 cycle as opposed to 2^128 cycles. Who knows, maybe someone acan come up with some uses for such capabilities in graphics.

In summary the end of science is nowhere near, not even in the horizon as far as I can see. Every discovery sets of dozens more.

[Mod]

For #2 , I also want to know what will financially happen to the companies, not just what the far future technology will be. Will the companies bankrupt ? Will they produce low quality products, so that we may replace it more often ? Something else ?

[Harnagel]

-Prices will reflect cost of research. If cost of research becomes too expensive for the market to justify purchasing the new product then yes those companies will go under. We are already seeing a decrease in the overall demand for CPUs due to the fact that there isn't much use for the newer speeds for the average user. So it also depends on whether we can find uses for these technologies that consumers are willing to pay for. Thus software is a big issue.

-I don't think they will get away with just charging more for less improvement.

-I personally don't think there is an "end" to the improvement in speed in the forseeable future. There will be cycles however, such as our current downturn.