Replies

From

Sony Cell CPU to deliver two teraflops in 64-core config

*********************************

Original URL: http://www.theregister.co.uk/2003/11/05/sony_cell_cpu_to_deliver/

Sony Cell CPU to deliver two teraflops in 64-core config

By Tony Smith (tony.smith at theregister.co.uk)

Published Wednesday 5th November 2003 12:11 GMT

'Cell', the massively parallel processing chip currently being designed by Sony and IBM, will scale from single-chip systems through to entire server rooms packed with thousands of them, Sony's executive deputy president Ken Kutaragi told attendees of the company's Transformation 60 conference yesterday.

That's always been the goal, of course, since Cell was first announced (http://www.theregister.co.uk/content/3/17511.html) back in March 2001. But yesterday Kutaragi put some numbers onto the chart.

A four-core chip home server system will be able to deliver one billion floating-point operations per second, apparently. Move up to a 32-core chip - in, say, a blade server module - and you'd get 32 gigaflops of processing power, while a 64-core slab of silicon inside a rack-mount unit doing graphics work would churn out two teraflops, according to Kutaragi's presentation foils.

Ultimately, Kutaragi suggested, we'll see 16 teraflop supercomputer 'cabinets' and one petaflop (a million billion flops, in other words) server rooms - the latter delivering enough raw power for true AI systems, he said.

Kutaragi likened a single Cell chip to IBM's 32-node RS/6000-based chess supercomputer Deep Blue. The exponential scaling rate suggests Cell really doesn't come into its own until you use lots of them together.

That's certainly the design philosophy: "With built-in broadband connectivity, microprocessors that currently exist as individual islands will be more closely linked, making a network of systems act more as one, unified 'supersystem'. Just as biological cells in the body unite to form complete physical systems, Cell-based electronic products of all types will form the building blocks of larger systems," was how Kutaragi described the Cell concept back in 2001.

Since it takes 2200 PowerPC 970 chips - aka the G5 - to yield (http://www.theregister.co.uk/content/61/33780.html) just over ten teraflops - much the same as you get from 2000 Athlon 64s (http://www.theregister.co.uk/content/archive/31994.html) - getting similar performance out of just 64 Cell cores is impressive, if Sony and co. can deliver.

Right now, they're just a little way past half way through the five-year Cell research project, so they have a few more years yet to demonstrate the device in action. We'd expect the successors to today's chips to have got a little closer to those kinds of figures by 2006, but not that close.

Others are not so far behind. ClearSpeed's recently announced CS301 chip, for example, can deliver 25 gigaflops peak, the company claims. The CS301 is a 64-way parallel processing co-processor designed to work alongside an x86 or other general purpose CPU. The downside is that it's expected to cost over $1000 per chip. We suspect Sony and IBM are aiming for something a little more mass-market. ®

Reaching back further:

Sony, IBM, Toshiba team on broadband supercomputing CPU

************************************************

The Register » Channel »

Original URL: http://www.theregister.co.uk/2001/03/12/sony_ibm_toshiba_team/

Sony, IBM, Toshiba team on broadband supercomputing CPU

By Tony Smith (tony.smith at theregister.co.uk)

Published Monday 12th March 2001 11:47 GMT

Sony, Toshiba and IBM are to co-operate on what they're calling a "supercomputer on a chip" and which to us sounds like the successor to the Emotion Engine processor currently driving the PlayStation 2.

Codenamed Cell, the new chip will form the basis of a massive $400 million, five-year R&D project. The goal of that research: to create a chip that not only goes like the proverbial off a shovel, but is designed specifically for high speed networks - and multi-processing across them.

Cell will utilise IBM's most up-to-date processor technologies - "copper wires, silicon-on-insulator transistors and low-K dielectric insulation" - all done at 0.1 micron (and probably a lot smaller by the time the thing ships) and fabbed on 12in wafers. It will "deliver teraflops of processing power". A 'flop' is a floating-point maths operation, and one teraflop is one thousand billion flops.

The trio haven't said much more about the chip, other than the tangential references to supercomputer performance and enabling the chip for a world of ubiquitous broadband Net connections. Ultimately, the part will power a wide range of consumer devices.

All that suggests we're looking at a general purpose microprocessor running at high speeds, enabled to process high throughput data streams coming in from all these wide, fast pipelines the partners expect devices containing Cell to be connected to. It's also expected to be a low-power processor. All that sounds like a cross between Emotion Engine, PowerPC and DSPs.

There's an irony here in that IBM fell out with its PowerPC partner Motorola when the latter developed the AltiVec vector processing system to allow the PowerPC to operate as a kind of programmable DSP. That's no use to us, complained IBM, and the two promptly parted ways.

IBM has clearly changed its mind about this kind of thing. It's certainly churning out Motorola-designed AltiVec-enabled PowerPC 7410 CPUs (aka G4) for Apple, and it may also have licensed AltiVec for the PowerPC-based processor it designed for Nintendo's GameCube console, which is due to ship later this year. And given Cell's emphasis on processing broadband-delivered multimedia data, it's not hard to imagine that IBM is working on something very similar to AltiVec for the new chip.

The chip's codename also gives a clue to its functionality. Said Ken Kutaragi, head of Sony's Computer Entertainment division: "With built-in broadband connectivity, microprocessors that currently exist as individual islands will be more closely linked, making a network of systems act more as one, unified 'supersystem'. Just as biological cells in the body unite to form complete physical systems, Cell-based electronic products of all types will form the building blocks of larger systems."

That suggests that multi-processing support is being designed into Cell from day one, doing all sorts of clever stuff like reading any chip's data and instruction caches, NUMA-style - a feature already on Motorola's PowerPC roadmap. And IBM has a keen interest in NUMA, thanks to its purchase of Sequent.

Kutaragi's spiel also implies some degree of clustering too. Again, that's a technology IBM has been poking its nose into a lot of late, allowing it to flog huge Linux-based supercomputing clusters to big name corporate clients.

But before everyone starts getting too excited by all this, here's one other point to note: IBM has been here before. The Sony-IBM-Toshiba alliance is uncannily reminiscent of the Apple-IBM-Motorola alliance. One was formed to create an Intel-busting next-generation computing platform and the other launched the PowerPC.

SIT seems less overtly anti-Wintel than AIM was, the beating-up-on-Intel aspect isn't absent. Kutaragi has talked about beating Intel before.

Whatever, AIM was unable to unseat Intel, and ultimately the partners ended up squabbling and chasing different markets. It's too early to say the same thing will happen to SIT, but five years is a very long time in the computer industry. A lot can - and will - happen between now and Cell's mass production and implementation. ®