Optimised, standards-based support for powerful new architectures
The latest oneAPI and AI 2023 tools continue to empower developers with multiarchitecture performance and productivity, delivering optimised support for Intel’s upcoming portfolio of CPU and GPU architectures and advanced capabilities:
4th Gen Intel Xeon Scalable Processors (formerly codenamed Sapphire Rapids) with Intel
Advanced Matrix Extensions (Intel AMX), Quick Assist Technology (QAT), Intel AVX-512, bfloat16, and more
Intel Xeon Processor Max Series high-bandwidth memory
Intel Data Center GPUs, including Flex Series with hardware AV1 encode and Max Series (formerly codenamed Ponte Vecchio) with datatype flexibility, Intel Xe Matrix Extensions (Intel XMX), vector engine, XE-Link, and other features
Existing Intel CPUs, GPUs, and FPGAs
The tools deliver performance and productivity enhancements and also add support for new Codeplay plug-ins for NVIDIA and AMD that make it easier than ever for developers to write SYCL code for non-Intel GPU architectures. These standards-based tools deliver choice in hardware and ease in developing high-performance applications that run on multiarchitecture systems.
What’s new in the 2023 oneAPI and AI tools?
Compilers & SYCL support
Intel oneAPI DPC++/C++ Compiler improves CPU and GPU offload performance and broadens SYCL language support for improved code portability and productivity.
Intel oneAPI DPC++ Library (oneDPL) expands support of the C++ standard library in SYCL kernels with additional heap and sorting algorithms and adds the ability to use OpenMP for thread-level parallelism.
Intel DPC++ Compatibility Tool (based on the open source SYCLomatic project) improves the migration of CUDA library APIs, including those for runtime and drivers, cuBLAS, and cuDNN.
Intel Fortran Compiler implements coarrays, eliminating the need for external APIs such as MPI or OpenMP, expands OpenMP 5.x offloading features, adds DO CONCURRENT GPU offload, and improves optimisations for source-level debugging.
Intel oneAPI Math Kernel Library increases CUDA library function API compatibility coverage for BLAS and FFT; for Sapphire Rapids, leverages Intel XMX to optimize matrix multiply computations for TF32, FP16, BF16, and INT8 data types; and provides interfaces for SYCL and C/Fortran OpenMP offload programming.
Intel oneAPI Threading Building Blocks improve support and use of the latest C++ standard for parallel sort, offers an improved synchronization mechanism to reduce contention when multiple task arena calls are used concurrently, and add support for Microsoft Visual Studio 2022 and Windows Server 2022.
Intel oneAPI Video Processing Library supports the industry’s only hardware AV1 codec in the Intel Data Center GPU Flex Series and Intel Arc(tm) processors; expands OS support for RHEL9, CentOS, Stream 9, SLES15Sp4, and Rocky 9 Linux; and adds parallel encoding feature to sample multi transcode.
Analysis & Debug
Intel VTune Profiler enables the ability to identify MPI imbalance issues via its Application Performance Snapshot feature; delivers visibility into Xe Link cross-card traffic for utilisation, bandwidth consumption, and other issues; and adds support for 4th Gen Intel Xeon Scalable Processors (Sapphire Rapids), Max Series (Ponte Vecchio), and 13th Gen Intel Core processors.
Intel Advisor adds automated roofline analysis for Intel Data Center GPU MAX Series to identify and prioritize memory, cache, or compute bottlenecks and understand their causes and delivers actionable recommendations for optimising data-transfer reuse costs of CPU-to-GPU offloading.
AI and Analytics
Intel AI Analytics Toolkit can now be run natively on Windows with full parity to Linux except for distributed training (GPU support is coming in Q1 2023).
Intel oneAPI Deep Neural Network Library further supports the delivery of superior CNN performance by enabling advanced features in 4th Gen Intel Xeon Scalable Processors including Intel AMX, AVX-512, VNNI, and bfloat16.
Intel Distribution of Modin integrates with new heterogeneous data kernels (HDK) solution in the back end, enabling AI solution scale from low-compute resources to large- or distributed-computed resources.
Beta additions for Intel Distribution for Python include the compute-follows-data model extension to GPU, data exchange between libraries and frameworks, and data-parallel extensions for NumPy and Numba packages.
Rendering & Visual Computing
Intel oneAPI Rendering Toolkit includes the Intel Implicit SPMD Program Compiler runtime library for fast SIMD performance on CPUs.
Intel Open Volume Kernel Library increases memory-layout efficiency for VDB volumes and adds an AVX-512 8-wide CPU device mode for increased workload performance.
Intel OSPRay and Intel OSPRay Studio add features for multi-segment deformation motion blur for mesh geometry, primitive, and objects; face-varying attributes for mesh and subdivision geometry; new light capabilities such as photometric light types; and instance ID buffers to create segmentation images for AI training.
Why should you use oneAPI?
With 48% of developers targeting heterogeneous systems that use more than one kind of processor*, more efficient multiarchitecture programming is required to address the increasing scope and scale of real-world workloads.
Using oneAPI’s open, unified programming model with Intel’s standards-based multiarchitecture tools provides freedom of choice in hardware, performance, productivity, and code portability for CPUs and accelerators. Code written for proprietary programming models, like CUDA, lacks portability to other hardware, creating a siloed development practice that locks organisations into a closed ecosystem.
*Evans Data Global Development Survey Report 22.1, June 2022
Buy the Intel oneAPI 2023 Toolkits with Priority Support
Alfasoft is an Intel Software Elite Reseller. We can support you with Intel oneAPI licensing advice and discounts. Purchase the Intel oneAPI 2023 Toolkits with Priority Support to gain access to private, dedicated Intel engineer support, access to earlier version, plus many other benefits.
For questions, please get in touch with Alfasoft via email or call us.