英特爾釋出2023版oneAPI多處理器軟體開發工具
英特爾（Intel）針對不同的處理器架構，釋出2023版本的oneAPI開發工具，新版oneAPI支援第四代Intel Xeon Scalable處理器、Xeon CPU Max系列處理器和Data Center GPU等產品，官方提到，新oneAPI不只獲得效能和生產力提升，也增加對新Codeplay套件的支援，開發人員可以更簡單地針對非英特爾GPU架構編寫SYCL程式。
因為有48％的開發者需要針對不同處理器的異構系統開發應用程式，因此需要更高效的多基礎架構程式開發，來滿足工作負載範圍和規模不斷擴大的需求。oneAPI提供開放且一致的程式開發模型，讓開發者針對各種CPU和加速器，擁有效能、生產力和程式碼可移植性的自由選擇靈活性。
oneAPI內含完整的英特爾2023開發人員工具，包括整套最新的編譯器、函式庫，還有分析和移植工具，也有最佳化人工智慧和機器學習的框架，可建構運作於CPU、GPU和FPGA多架構系統上的高效能應用程式。因為oneAPI提供標準工具，開發者能夠建構單一程式碼庫，快速實作應用程式並達成效能目標，進而更專注於程式邏輯開發。
新的oneAPI可讓開發者使用英特爾硬體進階功能，除了Xeon Scalable處理器、Xeon CPU Max的AMX、QAT、AVX-512和bfloat16等功能，也提供英特爾Data Center GPU資料型別彈性、向量引擎與矩陣擴充等功能。
oneAPI DPC++/C++編譯器增加了適用於Nvidia和AMD GPU的Codeplay軟體新套件，簡化編寫SYCL程式碼任務，並且使這些程式碼更簡單地在不同的處理器架構之間移植，官方表示，這對跨平臺開發提供了一個整合工具的統一建置環境。英特爾和Codeplay會從適用於Nvidia GPU的oneAPI套件提供商業優先支援。
而且由於開源專案SYCLomatic DPC++相容性工具，添加了100多個 CUDA API，因此開發者現在也更容易將CUDA程式碼移植到SYCL
New in Intel® oneAPI Base Toolkit 2023.0.0
Toolkit Level Updates
Intel® oneAPI DPC++ Compiler 2023.0.0
- Intel® oneAPI DPC++/C++ Compiler improves accelerated computing with more SYCL* 2020 features to boost CPU and GPU offload performance on the latest platforms.
- Intel® oneAPI DPC++/C++ Compiler adds support for 4th Gen Intel® Xeon® Scalable Processor (formerly Sapphire Rapids), and Intel® Data Center GPU MAX Series (formerly Ponte Vecchio).
- The compiler enhances OpenMP* 5.0 and 5.1 compliance and improves performance. New standard features have been added and enhanced for C20, C++20, and C++23,
Intel® DPC++ Compatibility Tool 2023.0.0
- The Intel® DPC++ Compatibility Tool (based on the open source SYCLomatic project) adds more than 100 new CUDA APIs migrating to the equivalent SYCL language and library functions including runtime, math, and neural network domains.
Intel® oneAPI Math Kernel Library 2023.0.0
- 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 Video Processing Library 2023.0.0
- The Intel oneAPI Video Processing Library supports the industry’s only hardware AV1 codec in the Intel Data Center GPU Flex Series and Intel® Arc™ processors; expands OS support for RHEL9, CentOS, Stream 9, SLES15Sp4, and Rocky 9 Linux; and adds parallel encoding feature to sample_multi_transcode.
Intel® Distribution for GDB* 2023.0.0
- Intel® Distribution for GDB* allows developers now to efficiently debug applications running on discrete GPUs using the L0 graphics drivers.
- Intel® Distribution for GDB* has been updated with more functional and stability improvements.
Intel® Distribution for Python 2023.0
- To reduce the default download size of the Intel Base Toolkit, Intel Distribution for Python* is now available via separate download.
Intel® VTune™ Profiler 2023.0.0
- Intel® VTune™ Profiler enables ability to identify MPI imbalance issues via its Application Performance Snapshot feature; and adds support for Sapphire Rapids, Ponte Vecchio, and 13th Gen Intel® Core™ processors
Intel® oneAPI Data Analytics Library 2023.0.0
- There are no new features for oneDAL in the 2023.0 release.
Intel® oneAPI Deep Neural Networks Library 2023.0.0
- Deliver production quality optimizations for Sapphire Rapids, Ponte Vecchio, DG2, and ATS-M with support for S8/S8 weights and activations allowing greater input influence on the outcomes on Intel SPR hardware with AMX, and wider operators - BF32 on Saphire Rapids CPUs and TF32 Pone Vechio GPUs yield more accurate inferencing
- Limited support for FP64 operators on PVC GPUs for high precision model deployment
- Deliver Graph API support to simplify integration to frameworks and extend optimization capabilities
Refer to Intel® oneAPI Toolkits and Components Installation Guide for Windows
評論