Million core CPUs

Post Reply New Topic

Another poll. When do you think we'll get to 1m (or more) cores in a single CPU and why?

I voted for 2030s because I'm assuming Moore's Law, aided by radical new chip designs, will continue until at least that time. Please remember that currently we're at quad-core CPUs.

Last edited by andy1976uk on 23 Jan 2007, 7:54 am, edited 1 time in total.

year 2007 = 4
year x= 1000000

assuming more cores will keep moores law going doubling after
1.5 years.

2^y * 4 = 1000000

x = y * 1.5 + 2007

2^18 = 262 144

18 (little less) * 1.5 + 2007 = 2034

Yup, that's pretty much my reasoning too. I just hope we can fully augment ourselves by then, as I don't particularly like the prospect of the human brain still taking 500,000 years to double in capacity in a world of 1m core CPUs evolving at Moore's Law.

I think by 2034 we'll have optical computers. With light you don't have the same limits as you do with electricity.

Or there's also the possibility of Quantum computing... but I don't know how soon that will happen. Will be incredibly fast, though.

Yeah and improved parallel processing. The human brain (neural network) uses parallel processing its much faster and requires
less processing units than a sequencal processing that is typical of
our PC world. Forexample the same transistors in a typical CPU
could be reconfigured* to execute a very simple program a thousand
times faster than the CPU could with a sequencal program (most
all typical programming). Part of the problem is translating a sequencal program into a parellel optimized program. I think part
of what Sony is counting on is a major break through in parellel programming that will give the PS3 a long life.

*reprograming logic forexample.

Why? Optical computing is the future.

I think you're overlooking the fact of CPU redundancy. The more CPU's that you add, the smaller the performance gain. I havn't kept up with emerging technologies lately, however quantum computing sounded really interesting when I first heard of the research that IBM was doing in that field back in 2K.

Back to parallel processing I mentioned if you could write a program to take advantage of 1 million typical processors cluster you could likely just as easy compile it into a FPGA's (reprogramable logic) and perhaps need 1/1000 of the transisters as the 1,000,000 cpu monster computer. Oh FPGA co-procesor cards are all ready available by the way. In the future if a programmable logic gate completely reconfigure in a single clock cycle it would make a device with about the same number of transistors as you present computer but easily 1000 times faster. The limitation as Fogman stated being their is not alot of programs written for highly parallell hardware.

"Sure, that's great man, but is it compiled for million cores? I don't wanna have quake 60 runnin like slag on my new rig dude"

The best use for multi-core systems is in a server application. Easiest situation to actually use multiple cores. Fast processor tech is always available to the server market first because they have the highest need and ability to use it... not to mention the funds.

BTW - Intel pledged a few months back to produce 80-core systems within 5 years.

@ idea of own personal Blade server or Mac X-Grid thingy (not the technical name).





....and Moore's Law continues unabated.

Do you know how many separate CPUs that will have?

Prolly on 1 to 4 chips... don't know

Moore said he never said 1.5 years but instead said 2 years. So
I am updating my estimate to 18 * 2 + 2007 = 2043
Its possible roadmaps will slow to 3 year cycles in the future to so
its reasonable to guess later than 2043.

we'll probably be using bio-computers before 1m cores.

Late 2020s.

Core may not even be the terminology by 2050.

Laser-based interdimensional transducers could be driving force behind Moore's law, by that time.