“Taking a software-first approach…”
The long-in-gestation Intel OneAPI project — a set of tools and libraries for building and deploying applications across diverse architectures including CPU, GPU and FPGA hardware — has finally landed, a year after first being teased.
The chip giant’s big reveal came at the Supercomputing 2019 event in Colorado this week, where Intel also revealed a range of other new products and services including a new 7nm GPU line for the data centre dubbed “Ponte Vecchio”.
The Intel OneAPI project aims to make it easier to build tools on top of scalar, vector, matrix, and spatial (SVMS) architectures deployed in CPU, GPU, AI, FPGA; it uses the Data Parallel C++ (DPC++) language: an evolution of C++ that allows code reuse across hardware targets, while permitting accelerator-specific tuning.
(The language is a cross-industry alternative to proprietary languages that’s a hybrid of C++ and SYCL, Khronos Group’s high-level programming model based on OpenCL.)
OneAPI will support both direct programming and API programming.
Intel has released a base tool kit to help programme hardware, along with six domain-specific toolkits, spanning high performance computing (HPC), the Internet of Things (IoT), debugging, analytics, rendering and OpenVINO – each designed to simply implementation of these domain specific applications varios accelerators.
A new line of GPUs meanwhile will use Intel’s Xe architecture, which the company has split into three differentiating marketing segments; consumer graphics, data centre and AI or HPC. Raja Koduri, GM of architecture, graphics and software said: “HPC and AI workloads demand diverse architectures, ranging from CPUs, general-purpose GPUs and FPGAs, to more specialized deep-learning NNPs…”
“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.”
The Ponte Vecchio graphics card will be manufactured using not just Intel’s 7nm processing wing, but will also use its ‘Foveros 3D’ high-performance three-dimensional integrated circuit and the firm’s packaging technology for high density interconnected chips the ‘Embedded Multi-die Interconnect Bridge’ (EMIB).
Although the company bills it as its first “exascale graphics card,” that kind of firepower would require multiple cards working together over very faste fabric.
Intel is working to install Ponte Vecchio graphic cards into the Aurora supercomputer system at Argonne National Laboratory, USA. Aurora will contain two 10nm-based Intel Xeon Scalable processors and six Ponte Vecchio GPUs. The Aurora system will eventually work at the exascale level which can compute a billion billion calculations a second or a thousand raised to the power of six (1018) operations per second.
It appears that two versions of the oneAPI are being released; an industry-wide facing one and an Intel beta that will give developers a portfolio of developer’s tools such as libraries, compilers and domain-focused toolkits.