Chip designer AMD has expanded the range of tools and supportOpens a new window for Linux-friendly features embedded in its system-on-chips (SoCs) x86 Zen architecture that combine central processing units (CPUs) and graphics accelerators, allowing developers to optimize processing speeds with open-source code.
The move confirms that US chip makers are taking an embedded route to meet faster processing demands in the data center and at the edge, driving workloads to servers and devices built using their silicon.
Not to be outdone, rival Intel is making an advanced version of its Xeon chips for data center servers compatible with Optane, a Dual Inline Memory Module (DIMM) architecture, to cut processing times by eliminating the software stack that stands between processor units and data pools.
On Monday, Intel revealed a pair of additionsOpens a new window to the product line built on the 20-year-old Xeon platform. The Cascade Lake advanced processor that combines 48 cores in each CPU was previewed ahead of next week’s formal Supercomputing 2018 event announcement.
Both Intel and AMD are working with specialists to help themselves and their OEM customer bases make more productive use of their chips and chipsets.
Partnerships Populate the Processing Field
Intel’s 2015 partnership with Boise-based Micron Technology led to Optane, which is made to run large, in-memory applications at low latencies. Together, the companies are continuing to develop the 3D Xpoint technologyOpens a new window , as well as operate a memory production facility in Utah.
AMD’s expanded line of code-ready chips and SoCs is a joint effort with Portland-based, design-automation software developer Mentor Graphics, a subsidiary of German industrial conglomerate Siemens.
The companies began work in 2014 on a support-and-tools suite called Sourcery Codebench for AMD’s G-Series and R-Series chipsets. The G-Series is made for applications requiring low power consumption and high configurability, while the R-Series permits faster processing of multimedia data feeds.
Users can license the full suite from Mentor, while AMD is making the same Lite version of Sourcery Codebench available for its EPYC and Ryzen chip families for data center CPUs and the Radeon Instinct graphics processing units.
Hybrid Applications Fuel Growth
When combined with Mentor tools, the architecture permits faster integration of AMD chips and chipsets with Linux-based operating systems and applications software. This is accomplished by tying together high-performance computing languages and interfaces to TensorFlow, and OpenCV libraries for machine learning and vision-based applications.
The tools enhance security, allowing developers to compartmentalize memory for encrypted data files and, claims Mentor, save money with the Integrated input/output.
In addition to emerging tech, use cases also are predicted for industrial processes and in medical imaging, applications that require both a high degree of processing capacity and rely intensively on graphics. Thanks to their ability to facilitate faster throughput, internet-of-things devices that require low-latency processing of large data flows also are high on the list.
Use cases for the Embedded EPYC chips are in networking and storage, where chips built on Intel’s Xeon platforms dominate. In addition to the Cascade Lake advanced processor, Intel launched the E-2100, a scaled-down version called the E-2100 aimed at small and medium-sized business and emerging cloud-services operators.
AMD’s push into the data center got more traction this month when Cray, a Seattle-based maker of supercomputers for government and research use, signed a $146 million deal with the US Department of Energy to deliver a machine built with EPYC chips.
With a design that permits users to scale their computing with CPUs, graphics processors and accelerators from different makers, Cray’s Shasta architecture is being touted as a vehicle for enterprise applications.
Along with the competition that underpins the push among chipmakers to deliver faster processing, development is being driven by scaling-up interoperability: Both Intel and AMD are members of the Yocto Project, a vendor alliance dedicated to standards-setting for embedded Linux.