IBM revealed the fresh details today at a chip conference in San Francisco.
“There’s nobody looking at anything like this. We have a more highly integrated chip that is multi-core and we are increasing the frequency – we are turning up both knobs at once when the industry is going the other way and turning [the frequency] knob down,” Bernie Myerson, chief technology officer in IBM’s systems group, told the FT.
IBM claims to have made major performance gains by stretching and squeezing silicon and using insulation techniques.
Power6 fits into IBM’s eCLipz (enhanced Core logic for iSeries, pSeries and z Series) project that will create a more common processor architecture for the company’s vast server line. Future zSeries mainframe products will likely rely on an offshoot of the Power6.
The strength of IBM’s Power4 and Power5 processors helped resurrect a Unix server line that had failed to compete with Sun during the boom times. Over the past four years, IBM has gained back large amounts of Unix server market share and demonstrated that its homemade processors can outperform rivals on a wide range of benchmarks.
Of late, however, the Power5 chip seems to be showing a bit of age. IBM has struggled to boost the chip’s clock speed.
Competitors have seized on this weakness to knock IBM as best as they can and some of IBM’s foes claim the company has fallen up to 18 months behind on Power6 production. Internal IBM roadmaps once showed that the chip might arrive in 2006, although a mid-2007 delivery date now looks most likely.
Most competitors have decided to keep the GHz of their chips flat, while at the same time adding more processor cores to each chip. This method gives a chip more horsepower without increasing power consumption and heat levels at the same rate as in the past.
IBM has not provided the nitty-gritty details on how it plans to achieve the performance gains with Power6. Industry watchers will keep a close eye on IBM in the coming years to see if can really deliver on these promises.
