The world’s first LLVM-based, unified compiler implements C++, SYCL, and Fortran to improve performance and productivity for CPUs and GPUs.

Data-parallel Python technology enables accelerated computing on CPUs and GPUs.

Intel® DPC++ Compatibility Tool now automatically migrates 90%-95% of CUDA cde to SYCL/DPC++.1

Advanced Development Productivity

Intel® VTune™ Profiler flame graph display improves the ability to visualize performance hotspots.

Intel® Advisor’s accelerator performance modeling estimates the benefits of offloading to a GPU before making code changes.

Expanded development environment support include deeper Microsoft VSCode integration, VS 2022 support, and WSL 2 for Linux development on Windows.

AI Training & Inference Performance Optimizations

Intel® Optimization for TensorFlow and Intel® Optimization for PyTorch accelerates deep-learning framework performance up to 10x over earlier versions.2

New Intel® Extension for Scikit-learn speeds machine learning algorithms over 100x on Intel CPUs compared to the stock open-source version.3

Intel® Neural Compressor is introduced to increase inference performance through post-training optimization techniques across multiple deep-learning frameworks.

Advanced Ray Tracing

New Intel® oneAPI Rendering Toolkit features including cone telemetry, auxiliary feature denoising, and FP16 support provide more robust shapes and shorten rendering times. Additional support for upcoming Intel® Xe GPU plus real-time denoising further improve final-frame, production-quality rendering.

Hardware Support

All toolkits are optimized to enable advanced features of the latest (and upcoming) hardware, including 12th Gen Intel® Core™ processors (codenamed Alder Lake) with AVX-VNNI, Sapphire Rapids with Intel® AMX, and upcoming Xe client and datacenter GPUs.