Intel's 80-core musclechip (and TRUE/FALSE/MAYBE)

Faster than a speeding IT Blogwatch: in which Intel crows about its new 80-core CPU prototype. Not to mention a new Boolean constant...

Ben Ames reports:

Following their march from standard processors to dual-core and quad-core designs in 2006, Intel Corp. researchers have built an 80-core chip that performs more than a trillion floating-point operations per second (TFLOPS) while using less electricity than a modern desktop PC chip ... 80 cores [on] a 275-square-millimeter, fingernail-size chip ... Intel ... [is] using the chip to explore new forms of tera-scale computing, in which future users could process terabytes of data on their desktops to perform real-time speech recognition, conduct multimedia data mining, play photorealistic games and interact with artificial intelligence.

...

Shrunk onto a single chip, that power would allow average consumers to use their PCs in new ways. They could use improved search functions on the vast amounts of digital media stored on home desktops, searching large photo archives for specific attributes such as all the shots where a certain person is smiling, or where that person is posing with a friend.

...

Running at 3.16 GHz, the new chip achieves 1.01TFLOPS of computation -- an efficiency of 16GFLOPS per watt. It can run even faster, but loses efficiency at higher speeds, performing at 1.63TFLOPS at 5.1 GHz and 1.81TFLOPS at 5.7 GHz. The processor saves power by shunting idle cores into sleep mode, then instantly turning them on as they're needed.

The delightfully-named Conrad Quilty-Harper adds:

Now that the Megahertz race has faded into the distance (we hear it was a myth), Intel is well and truly kicking off the start of a multi-core war ... It's not the first multi-core processor to reach double figures -- a company called ClearSpeed put 96 cores onto one of its CPUs -- but it's the first to be accompanied by the aim of making it generally available; an aim that Intel hopes to realize within a five year timeframe.

The long time frame is required because current operating systems and software don't take full advantage of the benefits of multi-core processors. In order for Intel to successfully market processors with CPUs that have more than say, 4 cores, there needs to be an equal effort from software programmers, which is why producing an 80-core processor is only half the battle. On paper, 80-cores sounds impressive, but when the software isn't doing anything imaginative with them it's actually rather disappointing: during a demonstration, Intel could only manage to get 1 Teraflop out of the chip, a figure which many medium- to high-end graphics cards are easily capable of. [Arrgh! It's "1 teraflops", dammit! The 'ps' stands for "per second".]

Here's Oliver Ryan with a perspective:

For the record, it's only been ten years (that's 1997) since the supercomputers first broke the teraflop barrier [Arrgh!]. The machine that did it was the ASCI Red at Sandia National Laboratories. Today's supercomputing champ is IBM's BlueGene/L, a machine capable of 280.6 teraflops.

Cyril Kowaliski has more detail:

[It's] an array of 80 "tiles" each containing a processing engine made up of a five-port router, two independent fully-pipelined single-precision floating-point multiply-accumulator (FPMAC) units, 3KB of single-cycle instruction memory, and 2KB of data memory. The two FPMACs are based on a Very Long Instruction Word-type design, much like Intel's Itanium. They have nine-stage pipelines and are able to provide an aggregate 16 gigaFLOPS of performance. And thanks to the five-port router, each "tile" can communicate with other tiles at up to 80GB/s.

Jagdeep Poonian points out:

A more technical examination of TeraScale is also available.

Loren Heiny cuts to the chase:

The multi-core processor doesn't run an x86 instruction set or support any of today's classic operating systems, but both of these points miss the more immediate value of a multi-core system such as this. There are a whole class of problems where a processor such as this--whether the main processing unit in a system or a complementary component--would be a significant boost to developers and end users. Just a few of these areas are in vision systems, synthesizing video, speech recognition, handwriting/shape recognition and search.

The graphics industry has already stepped in this direction for good reason and maybe with a more generic set of processors we'll see not only better looking images on our computers, but we'll see, hear, experience, and interact with our computers in ways that have not been practical to date. That's the key.

Dr. Spork is intrigued:

Ray tracing is embarassingly parallelizable, and while I'm no expert, two terraflops might just be enough calculating power to do a pretty good job at scene rendering, maybe even in real time. To think this performance would be available from a standard 65nm die that uses 65 watts... that really could make a difference to gamers!

John Murrell quips:

It's a good thing the cutthroat razor business can't escalate its blade count the way the semiconductor industry is starting to leapfrog on core count or a future generation of men would be trying to navigate some 50-edged monstrosity around their Adam's apples ... You could have had this kind of computing muscle in your home 10 years ago, of course, but you would have needed a really big rec room and a special deal with the power company

...

Just think of it -- with a teraflop [Arrgh!] in your hand, you could juggle artificial intelligence and instant video communications and photo-realistic games and multimedia data mining and real-time speech recognition and ... and ... Well, let's not get ahead of ourselves. First off, this is a lab model meant as a foundation for further work. Second, Intel still needs to figure out how to connect the chip with some memory. Third, developing a process for mass manufacturing is going to be darn tricky. And finally, here's a career tip for you young folks: Learn to write software for massively multicore chips.

Brian Lam thinks something smells funny:

Congrats on the teraflop [Aaargh!] chip, Intel, but wake us up when this baby is for sale. The Verdict: Vaporous CPU, solid PR move by Intel.

And it's a welcome return to Joe Duck, with this wise-quack:

One can’t help but think these speeds and power will soon break down the barriers between human minds and mechanical ones, leading to a revolution in thought the likes of which we may not be able to even imagine… without the aid of computer enhancements to our own brains! I just hope I can use my Circuit City coupons for a new, enhanced brain.

Buffer overflow:

Around the Net Around Computerworld Previously in IT Blogwatch

And finally... Things are rarely simply TRUE or FALSE

Richi Jennings is an independent technology and marketing consultant, specializing in email, blogging, Linux, and computer security. A 20 year, cross-functional IT veteran, he is also an analyst at Ferris Research. Contact Richi at blogwatch@richi.co.uk.

Copyright © 2007 IDG Communications, Inc.

Shop Tech Products at Amazon