Intel Corp. will begin producing its next-generation "Penryn" processors by the end of 2007, using greater power efficiency to push improved Core 2 and Xeon chips to speeds over 3GHz, the company said Wednesday.
The new chip family marks a crucial step in Intel's "tick-tock" product strategy, the company's schedule for delivering either a new chip architecture or smaller chip design every year, said Pat Gelsinger, general manger of Intel's digital enterprise group, during a press conference in San Francisco.
On March 5, Intel chief executive Paul Otellini apologized to investors for slacking off the development pace, and pointed to the tick-tock model as Intel's method for regaining some of the market share it has lost to competitor Advanced Micro Devices Inc. (AMD) in recent years. Now Intel plans to make a big splash in the market with the new product family as it begins production of six Penryn processors: dual-core and quad-core desktop chips, a dual-core notebook chip, dual-core and quad-core server chips, and a high-end server chip.
The Penryn design calls for shrinking Intel's current Core microarchitecture from chips using 65-nanometer feature sizes to 45nm. To prevent electricity from leaking between transistors packed so closely together, the chip will use novel "high-k, metal gate" materials to provide better insulation. IBM Corp. has also announced plans to use a version of high-k, metal gate design, but has not announced plans to bring it to market as fast.
The Penryn chip will also have better power management than previous Intel processors, with deeper sleep states than Core 2 Duo chips. Thanks to that efficiency, Intel plans to run its new chips faster than 3GHz for desktop and notebook versions, reversing an industry trend of scaling back the processor speed in order to add more cores without creating too much heat.
In other improvements, Intel will use 50 percent more on-chip memory in Penryn chips than Core 2 Duo, allowing them to hold more data on the chip instead of spending time and energy retrieving it from the PC's main memory bank. Dual-core Penryn chips will have 6M bytes of L2 cache while quad-core versions have 12M bytes. Intel said it also will speed Penryn front side bus speeds to 1600MHz, instead of the 1066MHz or 1333MHz options now available, granting up to a 45 percent improvement for high-performance computing applications like computational fluid dynamics.
In addition, Intel is adding a feature that takes into account a drawback of multicore chips, that they cannot boost performance unless a user upgrades from standard software to multithreaded applications. Penryn chips will recognize single-threaded applications and switch off power to the inactive core, speeding up the busy one.
Overall, the improvements will produce a dual-core chip that packs twice as many transistors into a space 25 percent smaller than Intel's equivalent 65nm processors, and uses equal or lower power, Gelsinger said. In full, Intel has 15 designs of 45nm chips now scheduled, he said. All the new Penryn chips will boast 20 percent faster transistor switching speed and lower leakage current than current chips.
Intel plans to upgrade its Core microarchitecture to a new "Nehalem" design in 2008, built first on 45nm architecture chips and then on a 32nm "Westmere" version in 2009, before giving way in 2010 to another new microarchitecture called "Gesher."
Nehalem will provide another leap in processing power by assigning two computing threads to each core, resulting in 16 threads for an eight-core chip that could deliver speedy performance for certain parallel applications.
The Nehalem design also calls for an "on-die integrated memory controller," a feature that AMD claims it used first in 2003 with its Direct Connect Architecture design that eliminates the front-side bus and integrates the memory controller. Likewise, some Nehalem chips will use on-chip integrated graphics processing, mirroring plans that AMD announced after acquiring graphics card maker ATI in 2006.
Gelsinger acknowledged the similarity but said that both companies had come to the same conclusion because they had seen the same handwriting on the wall.