Looking to meet the challenges of future high performance computing (HPC) systems, Intel says it will build the third generation of its Xeon Phi product family on a 10nm process technology. The chips, codenamed Knights Hill, will feature the company's Omni-Path Fabric technology, rather than the Infiniband interconnect.
According to Intel, Omni-Path will support line speeds of up to 100Gbit/s, with up to 56% less switch fabric latency than an InfiniBand based fabric. It will also use a 48 port switch chip, rather than the 36 ports supported by InfiniBand. More nodes per switch chip means fewer switches will be needed, simplifying the creation of HPC systems.
Intel has also created an 11 company alliance, called the Fabric Builders Program. The members, from the software and hardware communities, will look to develop Omni-Path based solutions.
Xeon Phi processors are built using Intel's Many Integrated Core architecture. The first generations of the Xeon Phi offered up to 61 x86 cores per processor, with clock rates of up to 1.2GHz. Using a multithreaded approach, each processor can run at up to 1.2TFLOPS.
* Meanwhile, the Czech based IT4Innovations supercomputing centre is building the largest Xeon Phi cluster in Europe. The supercomputer, being provided by SGI, will have 52,704 cores on 864 Xeon Phi 7120 coprocessors.