Thought Quad-core Was l33t? Try Quadra-vigen-core!

[rjhill1]

Intel managed to do it again...According to this BBC Article (http://news.bbc.co.uk/2/hi/technology/6354225.stm), Intel unveiled an 80-core processor capable of 1Tflops (FLoating point OPerations per Second). It is supposed to use 62W of energy, and each core can turn on and off as necessary to save power. It's also about the size of a thumbnail. The article also mentions the fact that just 11 years ago, when Intel first broke the 1Tflop barrier, it did so in a supercomputer that took up 2,000 sq. ft. Too bad it's not for commercial release...yet...

While normally I'm more of an AMD guy, I've really gotta hand it to Intel on this one. If I could afford to get my hands on that proc, you'd better believe I'd have one in each of my boxes. You'd be able to encode/decode/render even HD so much faster with that thing. And it's just amazing how far things have come, especially since the great wall of heat looked to spell the doom of x86 just a few years back (though admittedly this article doesn't mention whether or not it's an x86 proc). My congratulations to thee, Intel, this is engineering at its finest.

[rprice1]

Well, I didn't read too deep into the article when I was reading about it, but aren't the cores on the chip really simple? Just because you have 80 simple cores does that really exceed the power of something like a core 2 based processor? All I'm saying is that they need to show some benchmarks ;)

[rjhill1]

Well I watched Intel's demo video about the processor, and I really couldn't tell if it was x86 or not (which would kind of defeat the possibility of standard benchmark comparison), but from a wikipedia article estimating flops capability, the p4 clocked in somewhere in the hundreds of gigaflops range. Though they did say each core seemed to just consist of 2 floating point operators and a messaging router.

The teraflops ability would only really be useful to multimedia editors and/or scientific researchers, since that's when the floating point instructions would be used.

The main thing is the idea that they'll be able to scale this thing way beyond current limits, with them estimating they'll be able to dedicate cores to specific tasks, and each core would also be able to notify the others if it is becoming worn out or too hot to offload the work.

[cookster]

"It is going to require quite a revolution in software programming."

Yay for A.I....