News
News Categories

Meet Knights Landing, Intel's 2nd gen Xeon Phi coprocessor!

By Wong Chung Wee - on 17 Nov 2015, 3:18pm

Meet Knights Landing, Intel's 2nd gen Xeon Phi coprocessor!

A die shot of the Knights Landing coprocessor chip. (Image source: Intel)

Intel has launched its 2nd generation Xeon Phi coprocessor, codenamed Knights Landing. The company also officially announced the Omni-Path Architecture, its new, proprietary interconnect technology. These two announcements come under its Intel Scalable System Framework (Intel SSF). This framework is the company’s own architectural design to facilitate the building, configuration, as well as the maintenance of high performance computing (HPC) clusters.

In terms of hardware aspects, there’s the new Xeon Phi, and according to information obtained by PC World, the Knights Landing coprocessor will feature 16GB of multi-channel DRAM (MCDRAM). The benefits of the on-package MCDRAM modules are numerous; in comparison with DDR4, MCDRAM offers five times more bandwidth, and in terms of power efficiency, Intel claims the new memory is five times more power efficient as traditional GDDR5, and the new MCDRAM modules take up only one third the space of equivalent GDDR5 ones.

Intel Omni-Path host fabric interface (HFI) x8 adapter (Image source: Intel)

Working together with the new Xeon Phi coprocessors is the new interconnect Omni-Path Architecture technology. Intel claims its proprietary OPA fabric, built on open source and Aries fabrics technologies, will work with compatible switches and network cards, increasing link bandwidth up to 100Gbps. The new Xeon Phi coprocessors and select Intel Xeon CPUs will come with integrated Intel OPA controllers, eliminating the needs for separate interconnect adapters. In the future, Intel aims to make more components tightly coupled to further reduce the need for separate host interconnect components.

Intel Omni-Path 24-slot director switch (Image source: Intel)

According to AnandTech, the new Xeon Phi coprocessors are now being shipped to Intel’s first customers and developers. They are slated for launch in Q1 of 2016. With their general availability, together with compatible interconnect equipment, Intel aims to grow its SFF ecosystem and make it more cost effective for businesses and other relevant organizations to implement their HPC systems.

(Source: Intel, AnandTech, PC World)

Join HWZ's Telegram channel here and catch all the latest tech news!
Our articles may contain affiliate links. If you buy through these links, we may earn a small commission.