This morning on the WSJ’s D.Reside occasion, Intel formally unveiled its Nervana Neural Community Processor (NNP) household of chips designed for machine studying use circumstances. Intel has beforehand alluded to those chips utilizing the pre-launch title Lake Crest.
The expertise underlying the chips is closely tied to Nervana Methods, a deep studying startup Intel bought final August for $350 million. Intel’s NNP chips nix commonplace cache hierarchy and use software program to handle on-chip reminiscence to attain sooner coaching occasions for deep studying fashions.
Intel has been scrambling in latest months to keep away from being fully leveled by Nvidia. By refocusing on the rising AI market, the legacy chip maker certainly hopes to construct on its business connections to remain afloat. So far, Intel has been actively chasing a purpose of 100 occasions higher AI perforce by 2020.
The Intel Nervana NNP prioritizes scalability and numerical parallelism. The staff is promising sturdy bi-directional information switch. Utilizing a proprietary numeric format known as Flexpoint, Intel says it could actually obtain larger levels of throughput. And by shrinking circuit measurement, the staff notes it has been in a position to supercharge parallelism whereas lowering energy per computation.
In fact, each participant out there goals to construct chips that allow neural community parameters to be distributed throughout numerous chips at high-efficiency. We’ll know extra as soon as these chips hit the market.
As we speak’s announcement didn’t include benchmarking — in time. Intel says its chips will ship earlier than the tip of the yr. Fb has been supporting improvement by sharing technical insights with Intel.
Intel seems to have each intention of constructing a full product line round its Nervana NNP chips. A subsequent Xeon processor for AI has been rumored beneath the code title “Knights Crest.”