Intel OneAPI 2025.2

Toolkit Level Updates
- Improved cross-platform compatibility, performance, and efficiency with the Intel® oneAPI DPC++/C++ Compiler's enhanced SYCL interoperability with Vulkan* and DirectX* graphics APIs, benefiting applications like rendering and data visualization, among many others.
- C++ developers gain more efficient parallel algorithms programming coding options using Intel oneAPI DPC++ Library (oneDPL) expanded range-based APIs according to the upcoming C++26 standard, as well as the oneDPL set algorithms performance enhancement
- Data center application developers can now use the combined power of multiple GPUs for 2D and 3D non-batch FFTs using the new SYCL DFT API in Intel oneAPI Math Kernel Library (oneMKL).
- Easily migrate CUDA code to SYCL with the Intel DPC++ Compatibility tool automatically migrating 357 more APIs used by popular AI and accelerated computing apps.
- Streamline your debugging efficiency in Visual Studio with advanced thread filtering, focusing on active threads and simplifying error management, ideal for game development and real-time simulations. Enjoy regex validation, error handling options, and visual indicators for a more efficient and intuitive debugging experience.
- Enhanced debugging experience with the ability to set breakpoints on GPU kernels written in assembly from various inferiors, ideal for code offloaded to the GPU such as AI workloads. This includes debugging library calls from libraries like oneDNN and oneMKL, ensuring a more streamlined and efficient development process.
- Users of precision imaging apps can manipulate and align 32-bit floating point image faster on AVX-512 equipped Intel CPUs, both for apps using the Intel® Integrated Performance Primitives (IPP) Warp Perspective functions with Nearest Interpolation, or apps using OpenCV built with IPP
- Users of IPP Phase function to process 16-bit-integer image/signal data get several times faster performance on AVX2 and AVX512 capable CPUs
- Providing your app users the peace of mind of future-proof data security, by using Intel Cryptography Primitives Library new XMSS key/signature generation and enhanced SHA3 algorithms to protect data against classic and quantum computing threats
- Intel Cryptography Primitive Library's new HKDF algorithms empower developers to create short and strong secret keys to protect data and secure communication
- AI developers can achieve faster performance of DirectML-based AI applications on Intel client GPUs by optimizing for performance bottlenecks.
- AI developers can now optimize their AI applications running on Intel NPUs for performance by understanding the application frequency.
Intel oneAPI DPC++ Compiler 2025.2.0
New OpenMP 6.0 features:
- Added the stripe loop-transformation construct. Striping a loop or loop nest is useful to tune GPU offload performance.
- The nowait clause in target, target enter/exit data, and target update constructs can now take an optional Boolean argument to conditionally choose between asynchronous or synchronous offloading.
New OpenMP behaviors:
- Better support of debug info in OpenMP-outlined routines where some variables were previously reported as optimized away by gdb.
- For spir64 devices, a new command-line flag, -fopenmp-target-teams-default-vla-alloc-mode=malloc/wilocal (default: wilocal), was added to allow control over how local copies for variable-length arrays private to teams and distribute constructs are allocated.
- Added a new command-line flag, -fopenmp-target-loop-stride=local-size/global-size/one (default: local-size), to tune performance of spir64-offloaded OpenMP loops by controlling their loop stride.
Intel oneAPI DPC++ Library 2022.9.0
The Intel oneAPI DPC++ Library 2022.9.0
- Introduced additional range-based APIs for parallel algorithms. This work continues to improve alignment of oneDPL algorithm APIs with the upcoming C++26 standard.
- Improved performance of set algorithms.
Intel DPC++ Compatibility Tool 2025.2.0
The Intel DPC++ Compatibility Tool 2025.2.0
- Improved API coverage by supporting migration of 49 APIs (cuBlasLt/cuSparse/nvSHMEM/Texture/Graphics interop/ASM) used by priority apps, and other 308 APIs (nvSHMEM/Graph) with the available mappings.
- Support migration of CUDA Vulkan/DX Graphics Interop for 5 semaphores and 4 APIs.
Intel oneAPI Math Kernel Library 2025.2.0
- Introduces distributed SYCL DFT API with PVC support for computing 2/3D non-batch FFT on multiple GPUs.
- Introduced counter_engine_adaptor and (u)int8/16 types support in SYCL RNG Device API.
- The compiler team introduced device image compression in version 2025.0.1, and this feature is now enabled in oneMKL 2025.2 builds. It facilitates the reduction in size of SYCL libraries with large kernels by approximately 12% to 40% for both static and dynamic libraries on Linux, as well as dynamic libraries on Windows.
Intel Distribution for GDB* 2025.2.0
- Adds thread filtering options into VS and VSCode IDEs.
- Enhances CLI thread filter functionality.
- Adds group thread resuming feature.
- Enhances debugging of assembly GPU Kernels for oneDNN and oneMKL.
Intel VTune Profiler 2025.4.0
- Adds DirectML support via PIX markup device (GPU).
- Expands the reported set of NPU performance metrics (frequency).
Intel Advisor 2025.2.0
- Includes security and bug fixes.
Intel oneAPI Threading Building Blocks 2022.2.0
The Intel oneAPI Threading Building Blocks (oneTBB) 2022.2.0
- Links tbbbind with Static HWLOC for NUMA API predictability.
Intel Integrated Performance Primitives 2022.2.0
- Enables warp() IPP calls in OpenCV build with IPP.
- Enables AVX-512 optimization for ippiWarpPerspectiveLinear_32f_C1R.
- Improves performance for DFT/IDFT_16fc 72 and 216.
Intel Cryptography Primitives Library 2025.2.0
The Intel Cryptography Primitives Library 2025.1.0 optimized SM4 and SHA-512 hash family algorithms for Lunar Lake, Arrow Lake S and Clearwater Forest CPUs (optimization is based on the new SM4 and SHA512 instructions respectively).
- Crypto Multi buffer library was extended with Intel® AVX-IFMA implementation of ECDSA (Sign and Verify), public key generation, ECDHE over NIST p256r1 curve.
- Added support for HKDF, Hashed Message Authentication Code (HMAC)-based key derivation function as defined by RFC-5869.
- Added support for SHA3-224, SHA3-256, SHA3-384, SHA3-512, SHAKE128 and SHAKE256 hash algorithms as defined by FIPS PUB 202.
- Added support of Key and signature generations for the eXtended Merkle Signature Scheme (XMSS) algorithm.
Intel oneAPI Collective Communications Library 2021.16.0
The Intel oneAPI Collective Communications Library 2021.16.0 adds:
- SYCL graph support for Record and Replay for Allgather, Allreduce, and ReduceScatter.
- SYCL-based implementation of ring algorithm for Allgather.
- SYCL-based implementation for Broadcast.
- Multithread support for Allgather and ReduceScatter scale up impementation.
- Attribute in the communicator to specify blocking operations for CPU.
Intel oneAPI Data Analytics Library 2025.6.0
The Intel oneAPI Data Analytics Library 2025.6.0 contains:
- Better performance on linear regression with over 4k features
- Usage of higher SIMD instruction sets on AMD
- Added correlation distance metric
- Fixed a crash in GPU KMeans when the number of rows is large
Intel oneAPI Deep Neural Networks Library 2025.2.0
Intel oneAPI Deep Neural Networks Library 2025.2.0 has
- Improved performance on Lunar Lake and Arrow Lake integrated GPUs and Battlemage discrete GPUs.
- Improved performance on Granite Rapids CPUs.
- Extended flexibility of SDPA, MLP, and GQA fusion patterns with Graph API.

Toolkit Level Updates
The Intel HPC Toolkit contains updates included in the 2025.2 update for the Intel oneAPI Base Toolkit, plus the following:
- Intel oneAPI HPC Toolkit is a superset of the Intel oneAPI Base Toolkit so users enjoy all of the new features of Base Kit plus HPC Kit
- Enhance your programming efficiency with Intel Fortran Compiler's added F2023 features, offering the Selected_Logical_KIND intrinsic for advanced logical type handling and support for allocatable arrays of coarrays, perfect for optimizing parallel computing and complex data structures in scientific applications.
- Enhance your GPU offload performance and gain flexible control over offloading processes with OpenMP 6.0's new stripe loop-transformation construct and the nowait clause's optional Boolean argument to conditionally choose between asynchronous or synchronous offloading in the Intel oneAPI DPC++/C++ Compiler and Fortran Compiler benefiting high-performance computing and machine learning workloads.
- Boost your development with Intel Fortran Compiler's enhanced LLVM MemorySanitizer support for device code on Intel Data Center GPUs, ensuring robust memory safety and optimized performance for high-performance computing, scientific computing, data analysis, and a wide range of other applications.
- HPC developers and end users can achieve better resource utilization and performance with newly extended multi-threading capabilities for the full suite of collective operations, improvements to device-initiated RMA interoperability and performance, pinning support for NVidia GPUs, and other performance improvements.
- HPC application developers and end users have enhanced control over provider initialization with a new capability to set a list for I_MPI_OFI_PROVIDER.
- HPC developers can reliably implement MPI 4.1 functionality and expect stability and portability with updates to MPI 4.1 compatibility.
Intel oneAPI DPC++ Compiler 2025.2.0
New OpenMP 6.0 features:
- Added the stripe loop-transformation construct. Striping a loop or loop nest is useful to tune GPU offload performance.
- The nowait clause in target, target enter/exit data, and target update constructs can now take an optional Boolean argument to conditionally choose between asynchronous or synchronous offloading.
New OpenMP behaviors:
- Better support of debug info in OpenMP-outlined routines where some variables were previously reported as optimized away by gdb.
- For spir64 devices, a new command-line flag, -fopenmp-target-teams-default-vla-alloc-mode=malloc/wilocal (default: wilocal), was added to allow control over how local copies for variable-length arrays private to teams and distribute constructs are allocated.
- Added a new command-line flag, -fopenmp-target-loop-stride=local-size/global-size/one (default: local-size), to tune performance of spir64-offloaded OpenMP loops by controlling their loop stride.
Intel Fortran Compiler 2025.2
New Fortran 2023 features
- A data component whose type has a coarray potential subobject component may be an array.
- Intrinsic function SELECTED_LOGICAL_KIND has been implemented.
New Features
- The IEEE_SCALB (X, I) now permits the I argument to be of type real with the same kind type parameter as the X argument (Fortran 2018 interpretation)
- Option assume inline_cloc causes the compiler to generate inline code for the C_LOC function from the intrinsic module procedure ISO_C_BINDING. The default is assume noinline_cloc, which treats C_LOC as a call to an external function.
New OpenMP features
- The OpenMP 6.0 STRIPE loop transformation construct has been implemented.
- The OpenMP 6.0 NOWAIT clause now has an optional do-not-synchronize logical argument.
Intel MPI Library 2021.16
- MPI 4.1 compatibility updates
- OFI 2.0/PSM3 update
- Extends multi-threading capabilities for full suite of collectives
- Improved device-initiated RMA interoperability and performance
- Enhancement to restrict provider initialization to I_MPI_OFI_PROVIDER list
- Added pinning support for Nvidia GPUs and other Performance improvements