Light and CPUs

Page 2 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

deveraux

Senior member
Mar 21, 2004
284
0
71
Originally posted by: JediJeb
Originally posted by: tss4
Actually, both of you are right to a point. Both switching speed and internal propogation delays can limit the speed of a processor and thus its max frequency. However, what's interesting is that the ceiling on processor speed is now due more to a third problem. Heat dissipation. Theremal management in processors is now the biggest limitation to processor speed.

Now as far as the original post. Your 9 GHz is way off, as one of the other posters pointed out frequency of light is a simple calculation (speed of light in medium or vacuum / wavelength) which comes out to order of about 10^14. Perhaps they were refering to bandwidth of a type of optical fiber? Some fiber types have bandwidth distance products closer to that range and you could possibly confuse the two.

Now this does actually impose some interesting possibilities for speeding up computers for several reasons. One is that there would be no electromagnetic interference, so noise on increasingly dense circuits becomes less of a problem. Thermal dissipation is reduced. and lastly it is believed that we can make switches that operate faster than current electrical switches. Also , the bandwidth of light is far greater than electrical signals therefore the input and output to the processor could be faster.


Thats actually what I was thinking of, how far the pulse has to travel to trigger the switch, can't open or close faster than the signal can get there.

Yes right now the thermal problem is the BIG issue. Unless you submerge the boards in liquid nitrogen you are going to have trouble with overheating as you get faster.

Sorry, my mistake. After thinking about the issue a little more, internal distances would now actually play a bigger role than they did before as we're hitting ever higher frequencies. The reason I initially said that internal distances don't matter is because it is usually overrided by the bigger issue of transistor switching.

And yes, currently, thermal issue is at the top of the list of problems. Dual cores would be one way around it (well, for multithreaded apps anyway) and hopefully, Intel will figure out a way around their silly Prescott design and fix the damn leakage.