Intel’s new releases include advanced GPU architecture and OneAPI software stack

At Supercomputing 2019, recently, witnessed the unveiling of some latest Intel creations. These new releases stand for the company’s vision in the convergence of AI (Artificial Intelligence) and HPC (High-Performance Computing). All these additions have been made to Intel’s data-centric silicon portfolio along with an ambitious effort to move on from single-vendor, single-architecture programming models. All these maneuvers prove unchallenged leadership of Intel in its respective field.

What led to these developments?

Intel wanted to address the issue of increasing usage of heterogenous architectures in high-performance computing. The company announced a new discreet category of general-purpose GPUs customized and optimized for HPC and AI convergence.

Intel also plans to eliminate the hustle of multiple programming models across heterogeneous processing architectures including FPGAs, GPUs, CPUs and various other accelerators. To resolve this issue, the firm brought forth the oneAPI initiative that will lead to the creation of a simplified and unified programming model for app development.

OneAPI represents hours of labour put in by Intel engineers and promises to be a milestone in the evolution of present-day proprietary programming approaches for cross-architecture developer innovation and engagement.

oneAPI

This new initiative is set to define programming for the multi-architecture, AI-equipped world. It promises an open and unified programming experience to developers based on the architecture chosen by them without sacrificing the performance standards. It will also eliminate several complications like multiple programming languages, different workflows, various tools, separate code-bases, etc. oneAPI will play a key role in the preservation of investments along with support for all pre-existing languages while ensuring flexibility for developers in the creation of versatile applications.

Data-Centric strategy laying foundation of AI/HPC convergence

Intel’s portfolio of silicon is a mixture of diverse architectures working across a number of silicon platforms. Its new data-centric strategy is represented by the Xeon® Scalable processor that is currently empowering most of Top500 supercomputers on this planet. The Intel Xeon Scalable processors are the only ones with x86 CPUs with in-built AI acceleration optimized to evaluate the giant data sets in HPC workloads.

The Supercomputing 2019 witnessed the unveiling of a completely new category of standard GPUs on the basis of Intel’s X^earchitecture. Also named as the “Ponte Vecchio”, this general-purpose GPU is highly flexible, better performing, and discrete in nature. It is designed specifically for HPC modelling, AI training, and simulation workloads. The base for the development of Ponte Vecchio will be the 7mm technology by Intel, it will be the first X^e-based GPU specifically designed for AI and HPC workloads. It will also leverage Intel’s EMIB and Foveros 3D innovation packaging. It also features multiple technologies in-package inclusive of several intellectual properties, Compute Express LInk interconnect, and high-bandwidth memory.

In the words of maker

Raja Koduri, the chief architect, general manager, and senior vice president of Intel architecture, software, and graphics division, adds, “HPC and AI workloads demand diverse architectures, ranging from CPUs general-purpose GPUs and FPGAs to more specialized deep-learning NNPs, which Intel demonstrated earlier this month. Simplifying our customers’ ability to harness the power of diverse computing environments is paramount, and Intel is committed to taking a software-first approach that delivers a unified and scalable abstraction for heterogeneous architectures.”