Intel took the wraps off its new Xeon Phi chip, codenamed Knights Landing, during the International Supercomputing Conference in Leipzig, Germany. It’ll be capable of delivering over 3 teraflops of peak performance which is nearly three times as fast as its predecessor, Knights Corner.
The chip will feature a new high-speed fabric technology – Intel Omni Scale fabric – that’s said to address performance, scalability, reliability, power and density requirements to speed up the rate of scientific discovery.
Elsewhere, Knights Landing will use 16GB of stacked memory based on Micron’s Hybrid Memory Cube technology that’s said to provide 15 times more bandwidth than DDR3 and five times the bandwidth of DDR4 while being five times more efficient and consuming a third of the space. It’ll have more cores than its predecessor (61) although an exact figure hasn’t been released yet.
Knights Landing chips will be available for commercial system use sometime during the second half of next year and will arrive as both a standalone processor mounted in a motherboard socket and as a PCI-e card option. The latter option is described as a fast upgrade option although programming complexities and PCI-e bandwidth bottlenecks are absent in the socketed variety.
The first supercomputer to be powered by Knights Landing will be deployed by the National Energy Research Scientific Computing Center in Berkeley, California. Cori, as the machine will be called, will consist of around 9,300 Knights Landing chips and will be based on Cray’s interconnect.