Rupert Goodwins' Diary

Tuesday 25/10/2005 New processor companies aren't the commonest of creatures. The last one to make a big splash was Transmeta with its Intel-emulating chips whose primary attribute was low power without loss of speed.

Tuesday 25/10/2005

New processor companies aren't the commonest of creatures. The last one to make a big splash was Transmeta with its Intel-emulating chips whose primary attribute was low power without loss of speed. That never really took off — although the company's still around, it hasn't fulfilled its early promise.

This week sees another start-up come out of stealth mode. PA Semi has a range of devices promises, this time based on the IBM PowerPC architecture. What makes them special? Low power without loss of speed. Hm. True, it's very low power - a couple of watts producing as much CPU woof as chips taking ten or twenty times as much today — but don't expect to see so much as a sample for a long time.

You have to take PA Semi seriously, though. The CEO was the lead designer on the Alpha RISC processor — one of the best designs ever — and also worked on the StrongARM chip. His vice president of engineering produced AMD's Opteron architecture, while others have backgrounds with Xeon and Itanium. These people know how to make chips.

Can PA Semi be to IBM what AMD has been to Intel? It certainly strengthens IBM's story by bolstering the PowerPC line — system designers are always happier to have second sources for products; indeed, that's how AMD got into the x86 business in the first place, at IBM's behest. And while the technical claims for the new chips are breathtaking, on closer examination they're not too far adrift from what Intel claims it will be able to do with power reduction over a similar time period.

If IBM manages to keep PowerPC's momentum going - and it should, since despite Apple's flight into the arms of Intel the market is growing healthily — then there will be no shortage of takers for efficient, high-performing chips over the next five years. It should be an easier market to launch into than the x86, since there are fewer pressures to keep everything ultra-cheap.

In short: a good bet. If they can avoid spending all their money before getting to market, they'll be in with a shout of serious success by 2010.