researchers have developed a chip packing 80 individual cores, promising to bring supercomputing brawn to PCs within five to 10 years.
Intel plans to provide key details on its so-called Teraflop research chip in a paper at the International Solid State Circuits Conference taking place in San Francisco this week. The chip was first unveiled by CEO Paul Otellini at the company's developer conference last year.
Company officials stressed that the 80-core chip was not a future product per se, but rather a demonstration of various new technologies developed by the company that will eventually trickle into Intel products. The chip is based on new design and architecture that Intel says will allow it to deliver a "teraflop" of performance -- 1 trillion calculations per second, while consuming the same amount of power as current desktop processor.
Intel first achieved teraflop performance in 1996, with the ASCI Red Supercomputer that it built for Sandia National Laboratory. That computer featured 10,000 Pentium Pro processors dispersed in 100 cabinets within a 2,000-square-foot space, and consumed 500 kilowatts of electricity.
The new teraflop chip measures just 275 millimeters, slightly larger than a fingernail, while consuming 62 watts.
"It points the way to the near future when Teraflops-capable designs will be commonplace and reshape what we can all expect from our computers and the Internet at home and in the office," Intel Chief Technology Officer Justin Rattner said in a statement.
According to Intel, each of the 80 individual cores within the chip can be independently powered up or put to sleep depending on the application, allowing for efficient power consumption.
Each core sits on an individual tile, which features its own router operating alongside the computing core. That router connects to the other nearby cores, creating what Intel called a network-on-a-chip, as well as to a dedicated piece of memory stacked below each core.
The result is a modular design that can be scaled to create a chip with as many cores as befits the market and the manufacturing concerns Intel operates in.
"There's nothing magic about 80 cores, it could have been 40, could have been 100," said Jerry Bautista, director of Intel's terascale research program.
Intel's announcement comes as rival
Advanced Micro Devices
has publicly declared that getting locked into a race to cram more cores onto a chip is a dead end. AMD also is trying to bring supercomputing performance to desktop PCs, but the company is taking a slightly different approach by customizing the individual cores within the chip, most notably by adding a graphics processor into the mix.
AMD paid $5.4 billion for graphics chipmaker ATI to help it achieve that goal.
Intel officials noted that the teraflop chip is not about simply plunking more cores onto a chip. In fact, the company said it has conducted detailed simulations that show that after 16 cores, the performance actually starts to decrease.
To get more performance from chips with 32 cores or more, Intel said a chip needs to include new software instructions, cache improvements and smarter scheduling of the various tasks handled by each core.
"This is not just Intel declaring a new race for core count as the new metric," said Bautista. "Rather it's a requirement that a broad variety of technology needs to be addressed."