Prepare for Cell Processor hype onslaught

Just watching the US CNBC Squawkbox over here in England and they're going on about how amazingly powerful and revolutionary the new IBM Cell Processor is going to be. And if a product that isn't available yet and for which there is no new news available to get appearances on CNBC means that the PR engine has swung into full-on hype mode. Expect floods of blog postings about the chip and stories in all the news websites.

It's all PR hype though. We know the Cell is going to be the core of the Sony Playstaion 3 and there are possibly other applications for it in workstation accelerators. But don't expect to see it as widely used as IBM would like it to be. Firstly the Cell is not a cheap chip to manufacture. Some of the early hype made claims that by putting lots of simple processors together on one die they could make a very powerful chip at low cost compared to traditional solutions. This is just not the case though. The primary determiner in the manufacturing cost of chip is the die space it takes up and the Cell is as big as any high performance design. Even though the individual units of the Cell (SPE or Synergistic Processing Elements in IBM speak) are small and simple, when you jam 8 of them together (and add another true PowerPC processor to co-ordinate them) you end up with a big chip.

The other concern over the Cell is the ease of programming. Writing highly multi-threaded code is difficult anyway unless you have an easily seperable algorithm as your main computation task. But if you then take into account that the SPEs don't have access to main memory - only the 512KB they are bundled with - the whole business of co-ordinating data transfers to and from the SPEs adds an extra layer of complexity that any programmer would rather do without.

There is plenty of debate among game programmers over whether the PS3 offers as much useable processing power as the more conventional three-core PowerPC CPU in the Xbox 360 because the potention of the Cell will be so hard to realise. Being able to decode 48 MPEG2 streams simultaneously isn't that useful in most applications.