Tools
Explore new features and tools within Intel® products, communities, and platforms
83 Discussions

Freedom of Choice Across Accelerators with oneAPI

MaxTerry
Employee
0 0 2,590

This year’s Intel Innovation developer conference in San Jose, California, coincided with some big news coming out of the Open Source Summit in Bilbao, Spain. The Linux Foundation announced the formation of the Unified Acceleration Foundation (UXL), “a cross-industry group delivering an open standard accelerator programming model that simplifies development of performant, cross-platform applications.”

The UXL Foundation is an evolution of the oneAPI initiative. The oneAPI programming model allows for a single codebase to be deployed across multiple computing architectures, including CPUs, GPUs, FPGAs, and other accelerators. Intel will contribute its oneAPI specification to the UXL Foundation to help drive open source collaboration and development of a unified, cross-architecture programming model, free from proprietary software lock-in.

Rod Burns, VP of Ecosystem at Codeplay Software, an Intel company that has been at the forefront of oneAPI community governance, will chair the UXL Foundation Steering Committee. "The foundation members aspire to build the largest open ecosystem for accelerated computing. The initial contributions to the foundation will bring an existing open, standards-based platform with open governance. Our ultimate aim is to foster a multi-architecture and multi-vendor programming platform for all accelerators," he said.

At Intel Innovation, Andrew Richards, CEO of Codeplay, delivered a session with Dr. Valerio Pascucci of the University of Utah on how oneAPI provides a solution to the challenge developers and organizations face in harnessing the performance of diverse accelerator hardware for AI and other demanding workloads. As Richards noted, “48% of developers target heterogeneous systems that use more than one kind of processor or core.1” Yet the complexity of managing multiple toolchains specific to different architectures and vendors has impeded maximizing performance and productivity.

oneAPI is based on SYCL* from the Khronos Group, a C++ programming model designed to target heterogeneous architectures across a mix of processor types and vendor architectures. Richards, explained that “If you write your code in SYCL…you’re using industry-standard C++, and you then have the choice to run on any hardware you want.”

The oneAPI initiative offers an open source SYCLomatic tool that converts CUDA to SYCL code with comparable performance on Nvidia* GPUs. There is also an Intel implementation: the Intel® DPC++ Compatibility Tool, available as part of the Intel® oneAPI Base Toolkit. In addition, Codeplay recently released oneAPI for NVIDIA® GPUs and oneAPI for AMD GPUs (beta) – plugins that can be used with the Intel oneAPI Base Toolkit to build SYCL code and run on compatible Nvidia and AMD* GPUs.

 

oneAPI in Action – A Case Study

Dr. Valerio Pascucci of the University of Utah’s Scientific Computing and Imaging Institute shared how his team has employed oneAPI and SYCL in support of ecological research. The University of Utah and Lawrence Livermore National Laboratory’s oneAPI Center of Excellence used SYCL and Intel® software developer tools powered by oneAPI to develop scalable and portable massive image mosaic blending for the National Ecology Observatory Network (NEON) – a continental-scale observation facility funded by the National Science Foundation (NSF) and operated by Battelle.* NEON is designed to collect long-term open-access ecological data to better understand how U.S. ecosystems are changing.

The challenge NEON faced was that their aerial photography data displayed noticeable lines or divisions, known as seams, which arise due to variations in lighting conditions during the extensive data collection process. This occurs because the data collection spans vast geographic areas and takes place over long periods of time. As a result, the differing lighting conditions at different times and locations contribute to the appearance of these seams in the data. Such seams significantly impact the scientific value of the data since they introduce spurious artifacts in subsequent image analysis and interpretation of the underlying ecology. Dr. Pascucci’s team applied a Poisson iterative solver to develop an algorithm that enhances the raw image data, effectively removing false seams and providing data products that can be used for scientific investigation.

Valerio.png

 

 

Originally, this solution was created exclusively for Nvidia GPUs. However, Dr. Pascucci noted, its utilization needs to encompass a broad spectrum of hardware, both on-site and in cloud environments. Additionally, GPUs typically offer a more restricted RAM capacity compared to CPU computing instances. Consequently, incorporating a more flexible use of CPU/GPU cloud resources is key to achieve more extensive and realistic deployment where data processing can happen on the hardware resources available.

The solution? Cross-architecture Intel® software developer tools, powered by oneAPI which enable single-language and platform applications to be ported to (and optimized for) multiple single and heterogeneous architecture-based platforms. By adopting SYCL, the algorithm harnesses the computational power of diverse hardware architectures, ensuring consistent performance and reducing development efforts. Dr. Pascucci called attention to the development of a portable color-blending algorithm for photography data based on the Conjugate Gradient (CG) method and implemented it in the SYCL programming model to achieve scalability and portability on heterogeneous computing platforms. Dr. Pascucci noted that “the development was highly accelerated by the use of the Intel® DPC++ Compatibility Tool, which allowed us to translate the original CUDA code into SYCL kernels and MKL APIs.”

“Thanks to SYCL and oneAPI we now have a portable version of our blending algorithm which will allow easy deployment on different CPU and GPU based platforms used by the NEON team and their community.”

Watch a replay of this and other Intel Innovation sessions here!

Learn more about how you can maximize performance on the latest Intel hardware and take advantage of diverse accelerator architectures with Intel® software developer tools powered by oneAPI, and see more examples of industry and academic use cases.

 

1 Evans Data Global Development Survey Report 23.1

* Other names and brands may be claimed as the property of others. SYCL is a registered trademark of the Khronos® Group, Inc.

About the Author
Technology marketing leader with a passion for connecting customers to Intel value!