Intel debuts new release of its oneAPI multiprocessor software toolkit
Intel Corp. today introduced a new version of its oneAPI toolkit, which enables developers to build applications that can run on multiple types of processors.
Many companies’ technology environments contain multiple types of chips. For example, an organization can run its databases on central processing units and its artificial intelligence applications on graphics processing unit cards. Some companies also use more specialized chips, such as field-programmable gate arrays.
An application that can run on one type of chip is not necessarily compatible with another. Often, porting an application to a new processor architecture requires developers to make extensive code changes. The more chips a workload needs to support, the more code changes need to be made.
Intel’s oneAPI toolkit makes it easier to build applications that can run on multiple types of chips. According to the company, oneAPI reduces the amount of code that must be changed when porting an application from one processor architecture to another. The result is that developers can complete software projects faster.
The latest version of oneAPI that Intel debuted today has support for several of the company’s upcoming data center processors. It also offers improved compatibility with competing chipmakers’ silicon.
According to Intel, the new oneAPI release supports its upcoming Sapphire Rapids range of server CPUs. The company’s recently launched Xeon Max series is also supported. The latter product line includes CPUs optimized for high-performance computing systems, especially supercomputers.
Developers can now also use oneAPI to build applications for Intel’s upcoming line of data center graphics cards. The product line is taken up by the Max series, a chip with more than 100 billion transistors. It consists of 47 chiplets, or computing modules, optimized to run machine learning software.
Some applications require the ability to run on GPUs from not one, but multiple vendors. To facilitate the development of such workloads, Intel is extending oneAPI with a collection of new software plugins. The plugins were developed by Codeplay Software, an Edinburgh-based company that Intel obtain earlier in the year.
Developers use a programming language known as SYCL to write oneAPI applications. Code written in the language can run on multiple types of chips. According to Intel, the new oneAPI plugins will make it easier to write SYCL code that runs on Nvidia Corp. and Advanced Micro Devices Inc. graphics processing units can run.
Nvidia offers a software toolkit called CUDA with its GPUs. Companies use the toolkit to optimize the performance of their GPU-powered applications.
By default, CUDA-powered applications built for Nvidia chips are not compatible with other graphics cards. As a result, porting such applications to another chip manufacturer’s GPUs can involve a significant amount of work. To ease the task, Intel has updated oneAPI with features that will make it easier to turn CUDA-powered software into SYCL code that supports various types of graphics cards.
Intel also simplifies the development of GPU-based applications in other ways. As part of today’s update, the company added a feature that can automatically perform roofline analysis of workloads. Roofline analysis is a performance evaluation method that helps developers find ways to speed up their software.
“We are seeing encouraging early application performance results on our development systems using Intel Max Series GPU accelerators – applications built with Intel’s oneAPI compilers and libraries,” said Timothy Williams, the deputy director of Argonne National Laboratory’s computer science division. “For leadership class computer science, we appreciate the code portability benefits of multivendor, multiarchitecture programming standards.”
In addition to the new GPU features, the latest version of oneAPI also includes several other improvements. Intel has added improved support for OpenMP, an open source tool used by developers to build applications that can run on multiple processor cores simultaneously. Intel also improved oneAPI’s oneMKL component, making it easier to build applications that can perform complex mathematical operations.
