Intel announces 10nm Xeon Phi accelerators

Intel announces Xeon Phi processors for supercomputers produced at 10nm. The third generation of Xeon Phi will integrate a new version of the Omni-Path Architecture: an optical interconnect platform that is said to be faster than infiniBand.

During the Supercomputing Conference in the American city of New Orleans, Intel announced the successor to the upcoming Xeon Phi generation, codenamed Knights Landing. This successor will be called Knight Hill and should be produced at 10nm, while Knights Landing will be equipped with 14nm produced Silvermont cores.

When Knights Hill should appear is not known, but at least that will be after Knights Landing, which appears in the second half of 2015. Intel announced Knights Landing earlier this year. The Xeon Phi processors are accelerators for high performance computing.

Intel also announced that upcoming Xeon Phis will receive support for Omni-Path Architecture. This is an optical interconnect platform with throughputs of 100Gbit/s and a switch latency that would be 56 percent lower than current InfiniBand communication lines.

The Omni-Path Architecture will use a 48-port switch chip, where the maximum with InfiniBand is 36. The use of optical cables to connect the processors and other components in supercomputers should lead to speed gains and better scalability.