Sign up for our newsletter and get the latest HPC news and analysis.

Power6 at 5-ish GHz

IBM is to chip building as contrarians are to investing: going the other direction from the crowd.

Ashlee Vance at The Register has a piece on IBM’s upcoming 5-ish GHz Power6 chip:

At a time when rivals have given up on juicing their processors, IBM plans to crank the heck out of Power6. The dual-core chip due out by mid-2007 will come in “just shy” of 5GHz and flaunt double the performance of today’s Power5.

IBM feels that its Power6 is leading the industry, but The Register isn’t convinced:

But here we are in 2007 with IBM still sitting on dual-cores and a chip once meant to ship in 2006. In addition, IBM plans to take Power6 beyond 5GHz, while the entire industry scales back on pure speed in favor of numerous lower power cores that can digest more software threads.

And, by the way, neither am I. IBM is pouring resources into cranking up the frequency and managing power and heat on a path with a well-defined, physics-based end, while the rest of the industry has taken a left turn and is trying to find a solution that will have more of a future.

Not that I think cramming more cores into processors is the holy grail of computing. But I do think that it has a longer runway than cranking up the frequency. Of course, I don’t own a multi-billion dollar computer hardware company, and they clearly know more about that business than I do. Maybe they’ll surprise me.

Comments

  1. That’s got to be the dumbest move I’ve ever heard of. What’s worse is that IBM even made the BlueGene to have numberous low-clock-speed chips to deal with the performance issue. The hardware answer seems to be multicore CPUs alongside co-processors; this solution has low power requirements and a relatively low cost of adoption.

Trackbacks

  1. [...] You may recall that the chip was originally slated for 2006. This is the chip IBM previously revealed would approach 5 GHz clock rates. If that is indeed their target, it makes sense they could be having real yield problems in production. [...]

Resource Links: