The semiconductor industry is fond of quoting Moore’s Law, really Moore’s observation, that the number of transistors on an integrated circuit will double every two years.  In the past this has lead to faster and faster computers, but are we really getting faster computers today?

Think about it.  If you have a very fine mesh of transistors that you are going to push data into, then pull data out of after some algorithmic operation, the real performance of the system is based on (i) how much data you can push in and (ii) how fast you can clock it out.

Right now processors are hitting an I/O bottleneck that is limiting how much data can be pushed into and pulled out of your pile of transistors.  While the number of transistors on a chip continues to grow, data buses with 128 bit, 256 or more bits, even with multiplexing are unable to keep those transistors fed, meaning lit up with data.

Why have clock speeds plateaued at several gigahertz?  Clock speeds broke through megahertz level after level, but now face a thermal barrier.  The faster you clock, the more power you pump into the silicon and since you’re not increasing the area of the chip, then you’re putting more power into the same area and voila it gets hot.  If you increase the chip size to disperse the power, then fabrication yields drop.  Yes you can attempt some supercooling techniques, but that isn’t a viable economic answer to the question of computer performance.

The answer that the industry doesn’t want you to know is that we are at the end of significant performance improvements.  We are at the end of computer’s teenage years.  Computers have grown up.  The end of the performance renaissance is at hand.  You won’t know it for a few years because of industry marketing and our shift to smaller (smartphone and tablet) devices, but computers are not getting any faster.  Also the migration to the cloud which uses racks of computers operating in parallel will give the illusion of faster speed for a time.

The computer renaissance is over.  Answers will now only be solved by parallel processing techniques.  CPU speeds will not increase.  We are at the end of the trail and have reached the plateau.  For generations to come, this is as good as it will get.  The abacus computed for thousands of years without getting any faster, so too will its silicon descendants from here out.

One Response to “Clock speed wall and I/O bottleneck = end of faster computers”

  1. [...] Clock speed wall and I/O bottleneck = end of faster computers [...]

Leave a Reply



You may use these HTML tags and attributes: <a href="" title=""> <abbr title=""> <acronym title=""> <b> <blockquote cite=""> <cite> <code> <del datetime=""> <em> <i> <q cite=""> <strike> <strong>

© 2022 pdxmobile Suffusion WordPress theme by Sayontan Sinha