So, what is one to do with a teraflop desktop CPU?  Intel recently demonstrated their 80 core CPU that can do floating point math to the tune of a trillion operations a second while using a mere 63 watts of power.

Already, there are some projects to leverage to available computational power of modern video cards, which greatly exceeds that of the general purpose CPUs.  This is being done by some of the distributing computing projects, such as Folding@Home.  The current line of NVIDIA GPUs enables the use of 128 stream processors for this type of processing.  This is already more specific purpose cores than the 80 core demonstration by Intel, although Intel is planning on that being 80 general purpose cores. ATI was the first to be supported by Folding@Home. This led to some amazing performance gains.

Sure, all these new CPUs can be used for distributed computing where a single CPU will never keep up (by definition, since thousands will always perform better than one).  There will always be scientific and research projects that could benefit from very high end calculation performance (as defined at the time it's needed -- meaning the performance level of that definition goes up every day). 

Ok, so we'll need some of the new performance to process data off of our ever increasing network speeds (wireless or not). We'll always need more performance for our video games.  So long as any game still chugs along on the current highest end monitor resolutions, there will be room for improvement in games. 

Things like fingerprint recognition are here now.  Maybe in 5 years we can use the built-in webcams (that are hopefully better than 1.3MP resolution) for retina recognition in real-time. 

Microsoft Windows Vista tries to guess now what you'll do and when you'll do it.  Maybe in 5 years it'll be better at that. (I assume we'll likely still be using Vista, but probably SP2 or SP3 by then.) Perhaps other applications will join in on this predictive computing to reduce average latency of answering your questions or responding to your queries by guessing what you'll do next -- and doing it (of course, it won't be applied until you actually ask for it).

This reminds me of playing chess where each person tries to predict what the other will do and protect against it before it happens.  Speaking of chess, it's been nearly 10 years now since Garry Kasparov lost to Deep Blue.  The web site is still very much the same as it was back in 1997, too, which is odd. It talks a lot about "last year's" machine, which was actually Deep Blue 1996.  In 1997, Deep Blue could look at 200 million positions a second.  That was enough to just barely beat the highest ranked human chess player, who still holds that title. According to the Wikipedia entry, Kasparov called for a rematch that was denied and Deep Blue was retired (Kasparov is also retired now). The point of this, though, is that it's been 10 years.  Would there even be a possibility of a human player to beat a modern equivalent of Deep Blue anymore?  Aside from the special purpose chess chips, Deep Blue "only" ran at around 11 Gigaflops, at it's best.

Modern implementations only check a few million positions a second due to improved algorithms, but a teraflop-level CPU should be able to check significantly more.  At some point, a combination of truly deep brute force along with improving algorithms will make the computer chess engines nearly unbeatable.

This is one possible use of advanced computational ability.   Naturally, it's in games again, although not your traditional 3D game.  That said, if your 3D shooter AI dudes had the smarts of some of these chess engines, they'd be better able to train you for your first online encounter.

Outside of the scientific, research, and gaming communities, what can the average, normal computer user do with this sort of computational power?  Add in the fact that bandwidth could be increased by a thousand-fold in a similar time frame, there is enormous potential in the very near future for doing some very interesting applications.

What are those?  How do you even going about figuring out what those could be?  Can they be simulated now? What do you want to be able to do with your computer that you can't now?

Sure, I've posed similar questions before. If we really do get CPUs that perform at those levels in only 5 years, maybe sooner if competition and the need demand it, wouldn't you like to be ready for it?  Why wait for it to be here?  Chess algorithms of the past didn't wait for supercomputer to be able to work well.  Many were written such that as a speed of a computer goes up, the playing performance would go up because they'd be able to do more within the standard allotment of time.

It's all about thinking outside the box inside today's boxes.  ;)

Posted by Shane on February 13, 2007 7:54 PM | Permalink