Cell: the start of something big?

Recently, chip giants IBM, Sony and Toshiba jointly lifted the veil on their much-talked-about Cell microprocessor. Cell promises to be the biggest, baddest new chip on the block. It will end world hunger, cure disease, bring world peace-and make a killer videogame. What less would we expect from three such capable midwives?

Cell is interesting in part because it's an embedded processor. It's not some high-end workstation chip or the brains of the next missile-defense system. It's for toys and games. That three such massive conglomerates as IBM, Sony and Toshiba would spend so much time and effort on an embedded processor says a lot, I think, about where technology is headed. The innovation is not in the science lab anymore-it's in the living room.

First off, Cell is not one particular chip, it's the code name for a family of chips. The one announced recently is just the first of what the companies expect will be many. At its heart, Cell has a 64-bit Power processor designed by IBM, surrounded by eight 128-bit vector processors or accelerators. The whole thing runs at about 4 GHz in the lab, which is a respectable speed but one that Intel and others have already surpassed.

Note that Cell's central processor is a Power, not a PowerPC, design. Power is used in IBM's big iron and was around long before the PowerPC variation was invented. PowerPC is now predominantly an embedded processor (with a couple more chips going into Macintoshes), but this is the first time Power has been used in anything other than refrigerator-size servers. How times change.

The eight vector processors are fairly routine for a graphics-enabled device. They can handle either very large numbers or many smaller numbers at once. This makes them great accelerators for graphics routines, which are highly parallel and which sometimes require floating-point arithmetic. It's these vector processors, not the 64-bit Power core, that will make people's eyes pop when Cell-based products hit the market.

Those products will likely include Sony's Playstation 3, Toshiba DVD recorders and flat-screen TVs, and a gaggle of new satellite and cable decoders. Again, I think it's fascinating that so much expensive research and development has been expended on a chip for amusing the world's couch potatoes.

I proudly count myself in that group, by the way. I'm as much of a gadget junkie as the next guy. TiVo is the greatest invention since canned beer, as far as I'm concerned (another PowerPC-based box, by the way). But running spreadsheets, word processors and Web browsers has lost its charm. More to the point, those tasks don't tax today's computer designs. PC makers realized long ago that making PCs faster just doesn't appeal to most users; we upgrade our PCs more out of habit than necessity. Ah, but games? That's a different problem entirely.

Games-particularly graphically rich games-have an infinite capacity to eat up computing cycles. Computers can already calculate spreadsheets fast enough, but we're a long way from realistic 3-D graphics. There's a big difference between the special effects we see on movie screens and what we see on the Playstation. Animated movies aren't made in real-time, either. They're rendered over months on enormous (and enormously expensive) compute farms. Bringing that kind of graphic realism to a home device would take tons of computing power.

Or not. It might just take the right kind of processor. One designed from the outset for graphics performance, scalability and moderate cost. A processor that could grow and expand to handle somewhat different tasks or different applications; one that could spread across a big company's vast product line. Spread and expand like a virus, almost; a germ. Or a cell. This could be the start of something big.

Jim Turley (jturley@cmp.com) is editor in chief of sister publication Embedded Systems Programming.