The IEEE's International Solid-State Circuits Conference, which will convene next month in San Francisco, gives the world's top circuit designers their annual opportunity to one-up each other. Who'll walk away with bragging rights? Here's a sample:
IBM is presenting its zEnterprise 196 chip, which it claims is the fastest commercial microprocessor on the planet. The chip, built using IBM's 45-nanometer, silicon-on-insulator process, is the first to break the 5-gigahertz barrier. To make the best use of such high frequencies, IBM engineers had to come up with ways to improve the chip's consumption and distribution of power. The processor also includes a rarity in CPUs: a 30-megabyte high-speed dynamic RAM cache.
Researchers at Intel's Bangalore branch are obviously firm believers in the division of labor. They are set to introduce a Xeon processor with ten x86 cores, the most ever integrated onto a single server chip. The new Xeon, which is manufactured with a 32-nm complementary metal-oxide-semiconductor process, incorporates several design elements aimed at squeezing more calculations out of every watt of power spent. These include the ability to individually control the flow of power to each core, and a clock scheme for the cache and other shared components that lowers power consumption during idle periods.
Most Energy Efficient!
The Chinese Academy of Sciences in Beijing is showcasing a chip that gets it all done on an austere power budget. The Godson3B processor's 582 million transistors are capable of carrying out 128 billion floating-point operations per second while consuming just 40 watts. That translates to the best energy efficiency for a high-performance computing chip: 3.2 billion floating-point operations per watt.
How many transistors can you fit on a single die? Well, now we know that it's at least 3.1 billion. That's how many are on Intel's 18- by 30-millimeter Poulson Itanium processor. They're arranged into eight processor cores linked by nine layers of copper interconnects. The chip also features four shared caches and high-speed links that raise peak processor-to-processor bandwidth to 128 gigabytes per second.
What would a computer-chip Olympics be without a head-to-head competition between Intel and Advanced Micro Devices? Each chipmaker is presenting a graphics processor making claims to the title of most disparate elements integrated on a single die. Intel's Sandy Bridge processor boasts four x86 cores, an optimized graphics processing unit, a dual-channel memory controller, and a 20-lane bus interface, as well as a shared 8-MB cache. AMD's entrant, called Zacate, has 450 million transistors arranged into two x86 cores (each with a 512-kilobyte cache), a graphics and multimedia engine, and a high-bandwidth interface.