Omni Scale fabric
Two years ago, Cray Inc., of Mountain View, Calif., sold its interconnection technology to Intel for $140 million. Since then its team has been developing the next generation of that fabric, called Omni Scale. In turn, Cray recently announced that its next-generation supercomputer will be based on Intel's next-gen Knights Landing Xeon Phi.
At ISC-14 Intel has revealed that its next-gen Omni-Scale is a non-Infiniband optical fabric that will be integrated onto the Xeon Phi. It will be software-compatible with the last-generation True Scale Infiniband fabric, but will not require separate PCIe cards. However, it will include PCIe connectors for people who do not want integration.
Intel's next-generation Xeon Phi will top 3 teraflops by using the Silvermont architecture, on-package integrated silicon photonics fabric, and 3D hybrid memory cubes jointly developed with Micron.
Click here for larger image
Omni Scale fabric will also be available integrated on regular Xeon cores in future 14 nm processors, according to Wuischpard, and users will be able to upgrade from True Scale to Omni Scale photonics using existing optical connectors.
Lastly, Intel reconfirmed its commitment to educating programmers in the new techniques of coding for parallel processors. Its strategy is open parallel computing centers in major engineering universities.
"Intel has dramatically increased our investment in what we call the broader HPC ecosystem," Wuischpard told us. "In the last six months we have invested in over 30 parallel computing centers, typically grant-based collaborations with top researchers from around the world in two-year programs. We are now reconfirming that commitment by opening four to eight new centers per quarter for the foreseeable future."
— R. Colin Johnson, Advanced Technology Editor, EE Times