Sunday, February 06, 2005

The new Cell chip: can it succeed incrementally?

So far I don't get it. The The New York Times reports:
In a new volley in the battle for digital home entertainment, I.B.M., Sony and Toshiba will announce details Monday of their newest microprocessor design, known as Cell, which is expected to offer faster computing performance than microprocessors from Intel and Advanced Micro Devices. …

The Cell chip, computer experts said, could have a theoretical peak performance of 256 billion mathematical operations per second. With that much processing power, the chip would have placed among the top 500 supercomputers on a list maintained by scientists at the University of Mannheim and the University of Tennessee as recently as June 2002. …

"Our goal with the Cell is to be an order of magnitude faster," said Lisa Su, an I.B.M. executive in charge of technology development and licenses.

Many industry executives believe that because of its low cost, the Cell is a harbinger of a fundamentally new computing era that will push increasingly into consumer applications.

"I think it will aid in some of the convergence between consumer and corporate I.T. and this will accelerate amazingly from the consumer side," said Andrew Heller, a former I.B.M. processor designer who is now chairman of Heller & Associates, a consulting firm in Austin, Tex.

One area of wide speculation is whether Apple might become a partner in the Cell alliance in the future. Apple is already the largest customer for the PowerPC chip, and it would be simple for the company to take advantage of the Cell design. Several people familiar with Apple's strategy, however, said that the computer maker had yet to be convinced that the Cell technology could provide a significant performance advantage.
So why wouldn't Appple be interested? Something doesn't make sense.

Acording to The Register,
The 'cell' which gives the chip its name doesn't refer to the hardware, but to a virtual clump of software which roams the system looking for computing resources. The patent refers to a 'cell object' - program and data - and it can even roam across LANs or WANs, to find another Cell-based device.
and The Register (again)
Cell is designed to be a component in a massively distributed, global computing infrastructure. It's hardware specifically designed for 'grid computing'. A world full of Cell chips allows an entirely different infrastructure to take the place of today's transaction-based data centers. Software processes will scavenge the resources of the local Cell instantiation first, but if they find more execution resources over a local area network, or even on the other side of the world, they'll go and find them, and execute there.
This is a good idea, but it also says that Cell will succeed only if its vision of computing succeeds. Can it succeed incrementally? Is it also a superior (order-of-magnitude better) chip in a stand-along device?

No comments: