Now Available: Intel® oneAPI 2022.3 Tools Available

/in /by


#oneAPI #DPCpp #LLVM #Rendering #HPC #IoT

Enabling an Open, Multiarchitecture World

The newest update of Intel® oneAPI Toolkits and standalone tools is now available for direct download and/or use in the Intel® DevCloud for oneAPI. More than 30 tools are included in this release, each optimized to deliver improved performance and expanded capabilities for data-centric workloads.

Intel oneAPI Toolkits are purpose-built to optimize and accelerate cross-architecture and heterogeneous computing, delivering to developers open choice without sacrificing performance or functionality.

The toolkits provide compilers, languages, libraries, and analysis and debug tools that implement industry standards including SYCL*, C++, C, Python, Fortran, MPI, and OpenMP* as well as optimized versions of popular AI frameworks and Python libraries.

2022.3 Highlights:

Compilers

  • Intel® oneAPI DPC++/C++ Compiler adds more SYCL 2020 features to improve programming productivity on various hardware accelerators including GPUs and FPGAs and enhances OpenMP 5.x compliance.
  • Intel® Fortran Compiler adds Fortran 2008 and 2018 coarrays, DLLImport/DLLExport, DO CONCURRENT offload support, and -int and additional -check compiler options.

CUDA*-to-SYCL Porting

  • Intel® DPC++ Compatibility Tool supports more complete CUDA-to-SYCL code migration by adding support for CUDA 11.7 header files and CUDA runtimes and driver APIs including cuDNN, NCCL, Thrust, cuBLAS, and cuFFT.
  • The SYCLomatic Project expands Intel’s support of open computing with an open source version of the Compatibility Tool, which enables community collaboration to advance adoption of the SYCL standard.

Performance Libraries

  • Intel® oneAPI Math Kernel Library adds BLAS GPU device-timing support to ensure faster and easier detecting of exceptions and quicker recovery; improves portability and compatibility by extending OpenMP cluster offload capability to support the OpenMP 5.1 spec for LAPACK.
  • Intel® oneAPI DPC++ Library expands support of the C++ standard library in SYCL kernels with nine additional heap and sorting algorithms to simplify the coding of common functions.
  • Intel® oneAPI Video Processing Library includes the ability to provide extensive data about what is encoded, thereby opening up opportunities for quality improvement and algorithm innovation.

Analysis & Debug Tools

AI and Analytics

  • Intel® Extension for PyTorch* is updated to 1.12.100, includes automatic INT8 quantization, and adds operation and graph enhancements to improve performance across a broad set of workloads.
  • Intel® Optimization for TensorFlow* is updated to TensorFlow 2.9.1, includes performance improvements for bfloat16 models, and removes compiler requirement to enable oneDNN optimizations on Linux*-based Cascade Lake and newer CPUs.
  • Intel® Neural Compressor improves productivity with a lighter binary size, a new quantization accuracy feature and experimental auto-coding support, plus adds support for TensorFlow quantization API, QDQ quantization for ITEX, mixed-precision enhancement, DyNAS, training for block-wise structure sparsity, and op-type wise tuning strategy.

Rendering & Ray Tracing

  • Intel® Open Volume Kernel Library improves performance and memory efficiencies and adds support for VDB volumes (packed/contiguous data layouts for temporally constant volumes) and Intel® AVX-512 8-wide CPU device mode.
  • Intel® OSPRay supports primitive, object, and instance ID buffers as framebuffer channels, and face-varying attributes for mesh and subdivision geometry.
  • Intel® Embree supports the Intel oneAPI DPC++/C++ Compiler.

Get all the details >

Bookmark the oneAPI Training Portal – Learn the way you want to with learning paths, tools, on-demand training, and opportunities to share and showcase your work.