Intel oneAPI 2023.2 tools are now available. These extend and strengthen software development tools for open multivendor multiarchitecture accelerated computing.
Codeplay has also announced updates for plug-ins that seamlessly extend 2023.2 with SYCL support for NVIDIA and AMD platforms - bringing the enormous benefits of open, heterogeneous computing for your hardware. Learn the details in Codeplay’s blog about their updated plug-ins. Two other exciting annoucements from Codeplay, that relate to this 2023.2 release include: (1) uniquely valuable technology from the computecpp SYCL compiler project has been contributed to the open source DPC++ SYCL compiler project - bringing together these efforts to bring SYCL in the LLVM world - learn more in Codeplay's announcement on the completion of this Codeplay contribution of their advanced SYCL solutions into open source, and (2) Codeplay has introduced a oneAPI Construction Kit to further help hardware vendors looking to embracing multivendor multiarchitecture accelerated computing.
Standouts enhancements for 2023.2 release include:
- Faster Python for AI and HPC: First look (beta) at accelerated NumPy functions including Numba compiler support. Take a look – and share your experiences while you enjoy the performance boosts these offer for Python applications. We will extend these more in our 2024.0 release coming later this year.
- Faster and more accurate AI inferencing thanks to new capabilities that adds “missing values” (NaN) support during inference to streamline pre-processing and boost prediction accuracy for models trained on incomplete data.
- Faster oneDNN and graph compiler preview: Intel® oneAPI Deep Neural Networks Library (oneDNN) has numerous optimizations plus a new experimental Graph Compiler backend for the Graph API that enhance
- Image Denoising accelerates AI-based image enhancement on GPUs: Exciting new Intel® Open Image Denoise supports Intel, Nvidia and AMD GPUs--providing choice to ISVs, studios, and researchers for fast, high-fidelity AI-based image enhancement on a wide range of hardware.
- Performance enhancements: Numerous tuning improvements, and fixes for customer reported issues. Highlights include significant SYCL compilation speedups, math library performance improvements, and standard C++ parallelism routines for the Intel® Data Center GPU Max Series.
- SYCLomatic improvements: This popular open source tool that helps migrations from CUDA to SYCL has been enhanced with expanded support for CUDA APIs, FP64, and the latest version of CUDA.
- std::invoke parallelism: Intel® oneAPI Threading Building Blocks (oneTBB) algorithms and Flow Graph nodes now can accept new types of user-provided callables such as pointers to member functions and pointers to data members, resulting in a more powerful and flexible programming environment. This is accomplished by using std::invoke.
- DO CONCURRENT reductions: Intel® Fortran Compiler extends support for Do Concurrent Reduction, a powerful feature that can significantly improve the performance of code that performs reductions while also making it easier to write efficient and correct parallel code.
- Uninitialized Variable detection: Intel® Fortran Compiler LLVM memory sanitizer helps find errors related to uninitialized variables.
- Crypto XTS mode of SM4 algorithm: Intel® Cryptography Primitives Library multi-buffer library now supports XTS mode of the SM4 algorithm, benefitting developers by providing efficient and secure ways of encrypting data stored in sectors, such as storage devices.
- GDB 13: Intel® Distribution for GDB rebases to GDB 13, staying current and aligned with the latest enhancements supporting effective application debug and adds debug for Shared Local Memory (SLM.)
- Impressive performance for Intel® GPUs including ray-tracing support: Most tools and library updates include refined optimizations and functionality supporting Intel® GPUs. Additional articles highlighting this will be posted in the upcoming weeks.
I also will mention Velocity Bench. When GPU performance matters, we find that comparing apples to apples is challenging. A recent release of “Velocity Bench for GPU Offload Performance Data” at least helps establish codes that can be used and modified to help in comparison efforts. Such efforts to raise understanding through open community efforts to find cross-platform benchmarks is related to the goal of oneAPI to make all hardware more accessible to all applications.
oneAPI tools and plug-ins are making real the dream of an open future for accelerated and heterogeneous computing. The approach provides paths to open source support for hardware, and it also connects to closed source library options should they be preferred (e.g., cuBLAS for NVIDIA). In other words – the goal is “no excuses” access to performance for all vendors and all architectures (not just GPUs). Codeplay Software offers prebuilt plug-ins for the oneAPI tools to add support for NVIDIA and AMD GPUs.
Available Now: Intel® oneAPI Tools 2023.2
These programming development tools are available for download from Intel® Developer Zone, as well as via repositories and other channels.
The tools are also on the widely used Intel® DevCloud. Our experimental (beta) version of Intel® Developer Cloud to offer "PVC" (officially known as Intel® Data Center GPU Max Series) access has the 2023.2 tools (Base and HPC toolkits) installed as well. Keep in mind this project is fully beta - and has bumps (known issues are listed in instructions as well).
Join oneAPI to Help Create the Best Future Together
These tools and related plug-ins are key in our shared journey to foster an open multivendor and multiarchitecture future for computing.
The oneAPI initiative, and its open specification, are focused on enabling an open, multiarchitecture world with strong support for software developers—a world of open choice without sacrificing performance or functionality.