Intel’s teraflops chip uses mesh architecture to emulate mainframe

Research into competing architectures for the multicore processors of the future will take center stage this week when Intel Corp. demonstrates its Teraflop Research Chip–code-named Polaris–at the International Solid-State Circuits Conference in San Francisco. The 80-core chip crunches 1 trillion floating-point operations/second when running at a 3.2-GHz clock speed and consumes 62 watts, to yield a record 16 Gflops/watt. And by cranking the clock up to 5.6 GHz, the chip bested 1.8 teraflops–that’s 80 percent faster–albeit by increasing power consumption fourfold to 265 W, or 3.7 Gflops/W.