TeraFlop computing@home

in General Discussion edited January 2014
So, I was in my CS class a little while back, and the prof was going over basic terms: byte, hz, RAM, flop etc (c'mon, it was the first day of an intro. course!!! ).

He explained the difference between personal computers and supercomputers in terms of flops, which seemed to be his favorite metric. At the end of the little lesson, he said that at current rates of speed increases and architectural advances, he believed that PCs (generic term here!) would be capable of TeraFlop computing in about eight years.

Well, I wondered about that prediction. I googled around for a little while and from what I found, I think that cell (or some other multi core architecture) seems to be the only realistic and affordable way to achieve this. That said, I suspect he based his prediction on a simple expansion of Moore's Law (bearing in mind, of course, that it concerns number of transistors and thus that speed increases tend to be a by product rather than a goal...)

So what do others think? Is this an achievable goal within his predicted timescale? If not, will it arrive earlier or later? What sort of uses will such power have in a PC? Are we looking at systems that might be able to do something like real time ray-tracing? Photorealistic gaming? True speech recognition? In home Artificial Intelligence?

Would such a system require an evolutionary OS or a paradigm shift?

Or will such systems remain in the realm of science fiction for a while yet? (please be kind if you reply; I am something of a CS novice looking to learn as much as I can......)


  • Reply 1 of 6
    blablablabla Posts: 185member
    Depends on what you mean by TF.. is it 32 or 64 bit floating point, is it in real-world applicaiton or only theoretical peak performance, is it the general purpose CPU alone or the complete (including graphics) system. With the help of Altivec, the G5 got a theoretical peak performance of 32 GFlops. I dont think the fundamental task you are able to do with a computer would change that much if the speed was increased ~30 fold.

    Artificial intelligence research requires A LOT of computing power.. at least if you are talking about ANN/evolutionary computation. Its one of those funny fields where in most cases you got enough RAM, 32 bit FP is good enough, but you can't get enough computing power.
  • Reply 2 of 6
    Well, I guess he meant real world as opposed to theoretical peak, since those are by definition never achieved.
  • Reply 3 of 6
    wmfwmf Posts: 1,164member
    Today a PC can do over 10 GFLOPS, and performance has been increasing by about a factor of 10 every 10 years, so PCs should reach the TFLOP range in around 20 years.
  • Reply 4 of 6
    Well, iirc, Ken Kutaragi at Sony believes it will be possible within the next few years....maybe in the PS3 cell proc.
  • Reply 5 of 6
    yeah, I remember reading the ps3's future plan, they estimate the cell cpu that's gonna be one it will run at about 5 ghz. and that's for a gaming system I feel cell will be the future, and so will fiber optics. they say since fiber optics is so much faster then plain cables, you can hook up computer hardware with it, and utilize lasers that flash to send signals. sounds cool, and cool looking
  • Reply 6 of 6
    Heh, not just "a few years", but as soon as next year! Supposedly they are piecing together prototypes as we speak. ...but alas, maybe all of this is just a misinterpretation snowballing out of control. I believe the original claim was that PS3 will be 1000x the performance of a PS2, not necessarily 1000 GFLOPs. I also don't think 5 GHz was an actual real target, either. As it sits now, I think the current consensus expects around 2-4.
Sign In or Register to comment.