Intel Shows Off 80-core Processor

thejakebrain writes "Intel has built its 80-core processor as part of a research project, but don't expect it on your desktop any time soon. The company's CTO, Justin Rattner, held a demonstration of the chip for a group of reports last week. Intel will be presenting a paper on the project at the International Solid State Circuits Conference in San Francisco this week. 'The chip is capable of producing 1 trillion floating-point operations per second, known as a teraflop. That's a level of performance that required 2,500 square feet of large computers a decade ago. Intel first disclosed it had built a prototype 80-core processor during last fall's Intel Developer Forum, when CEO Paul Otellini promised to deliver the chip within five years.'" Update: 06/01 14:37 GMT by Z : This article is about four months old. We discussed this briefly last year, but search didn't show that we discussed in February.

IA64

[AvitarX]

I remember when IA64 was the next huge supercomputer on a chip 5 years off.

It didn't work out too well for Intel.

Deja vu all over again

[$RANDOMLUSER]

"Intel CEO promises to deliver magical new uber processor within five years".

Stop me if you've heard this one before...

core 2 duo has a higher transistor density?

[doombringerltx]

Intel used 100 million transistors on the chip, which measures 275 millimeters squared. By comparison, its Core 2 Duo chip uses 291 million transistors and measures 143 millimeters squared.

Maybe its just because I haven't had my morning coffee yet, or is that a typo?

Re:core 2 duo has a higher transistor density?

[Andy Dodd]

80 cores means there are probably quite a lot of on-chip interconnects between the cores.

Not only a dupe... but of an old story

[CajunArson]

It must be a really slow news day. From the dateline:

Published: February 11, 2007

Not to mention that Slashdot (even Zonk) Covered this LAST YEAR [slashdot.org].
But that's OK, I'm sure Slashdot gave insightful and cogent coverage of real events that actually matter to geeks on this site, you know, like the Release of a new major version of GCC [gnu.org]
Oh wait.... that (like a bunch of other actually interesting stories) would be in the aptly-named, sir not appearing on this website category due to it not making enough banner revenue.

Re:It may be known as "a teraflop", but...

[Anonymous Coward]

The measurement is "FLOPS". Floating Point Operations Per Second.

But that spells "FPOPS".

If we're going to be speaking strictly, get it right:

FLoating point Operations Per Second

AMD's response

[drgonzo59]

This is a nice move by Intel. I wonder what AMD's plans are...81 cores?

Besides, with most software being single-threaded I don't know if a consumer will immediately need more than 4 cores for a while. I can still see software companies trying to come up with ways to keep all 80 cores busy..."Well, they need at least 20 anti-virus processes, 10 genuine advantage monitors, and we'll install 100 shareware application with cute little icons in the task bar by default. There, that should keep all the cores nice and warm and busy -- our job is done!".

But in all seriousness, I would expect some extremely realistic environmental physical simulations (realtime large n-body interactions and perhaps realtime computational fluid dynamics)...now that's something to look forward to!

For the love of god...

[tomstdenis]

and all that is holy on this sacred Earth ...

This isn't a general purpose processor. Think "cell processor" on a larger scale. You wouldn't be running your firefox or text editor on this thing. You'd load it up and have it do things like graphics processing, ray tracing, DSP work, chemical analysis, etc...

So stop saying "we already don't have multi-core software now!!!" because this isn't meant for most software anyways.

Tom

Re:For the love of god...

[ciroknight]

"This isn't a general purpose processor."

...Yet. You're right in the fact that these cores are incredibly simplistic, so much so that they make DSPs look functional, but really what's going on here is a science project to develop the on-chip network, not to develop the CPU cores as much. Intel envisions lifting the networking component out of this design and applying it to various different cores, so that a general computing core can be mixed in with DSP cores and other "Application Specific Accelerator" cores.

So no, this model you're not going to be running Firefox or your text editor on (in fact, I doubt you even _could_ do this, these cores currently are very, very stripped down in their capacity to do work, to where they're basically two MACs tied to a small SRAM and a "network adapter"), but never-say-never, this style of chip is right around the corner.

Re:Frak everything, we're doing 80 blades

[jollyreaper]

I find this even funnier:

http://money.cnn.com/2005/09/14/news/fortune500/gi [cnn.com] llette/ [cnn.com]

Well shit, they should just rename the Onion to The Daily Prophet. Remember that little bit they did about Bush after the first time he was (s)elected, "Our Long National Nightmare of Peace and Prosperity Is Over?" http://www.godlessgeeks.com/BushNightmare.htm [godlessgeeks.com] Here it is with links to all the jokes that came true. Shit!

Re:cue

[benzapp]

Do you have any evidence of this? Are you saying that if I write an application to execute 16 threads simultaneously 16 different processors on a machine running Vista, that application will not see any speed increase over running that 16 thread application on an 8 processor machine?

Why would this be? And what is with the mac pro nonsense? Do you really think only apple makes 8 core machines?

Re:Older Story

[timeOday]

Sure would be nice to have a play with it once they have worked out how to program it...

It's very likely you can get one at Best Buy before they have worked out how to program with it. The fact is, current programming paradigms simply aren't suited to fine-grained parallelism - and in saying that, I don't mean to imply that such a paradigm can definitely exist. Sure there are many parallel research languages, but whether those could be adopted by mainstream programmers and used to achieve anywhere near linear speedup on mainstream applications is an open issue. Even the PS3, which is oriented to media applications which are relatively easy to parallelize, is getting little benefit from its measly half dozen computational units.

But with single-thread performance growth at a virtual standstill, Moore's law is going to result in exponential growth in the number of cores, whether or not we're ready to write software for them.

I wonder if we won't move towards a more "biological" paradigm - massive parallelism, but with massive redundancy and therefore inefficiency from a computational standpoint, but also robustness to hardware and software bugs.