Suggestions appear as you type. Use the up and down arrows to choose one and Enter to open it.

RAJA review

Free#31 of 39 in GPU Computing SDKs

Open-source C++ loop abstractions spanning CPU, NVIDIA, AMD and heterogeneous systems.

6.2/10Editor score
RAJA6.2 Visit RAJA

Reviewed by iTechGuides Editors · Editorial team · Updated Nov 2026

RAJA is an open-source C++ library from Lawrence Livermore National Laboratory for high-performance computing teams that need one application codebase across different CPU and GPU systems. Its policy types and templates separate loop bodies from execution methods, covering parallel loops, reductions, scans, atomic operations, sorting, data layouts, views, iteration spaces and compile-time loop transformations. Supported architecture targets include NVIDIA GPUs, AMD GPUs and heterogeneous CPU/GPU systems, with Linux, Windows, macOS and self-hosted deployment options.

The published model is straightforward: RAJA has an open-source free plan rather than paid tiers. Integrations include Umpire, CHAI, CMake, BLT, camp, CUB and rocPRIM, giving projects connections to memory management, portability utilities, build workflows and vendor-oriented parallel libraries. Execution policies span sequential CPU, SIMD, OpenMP CPU threading, TBB, NVIDIA CUDA, AMD HIP, OpenMP target offload and SYCL. This breadth suits organizations standardizing C++ loop code across varied accelerator environments, particularly mid-market and enterprise HPC teams.

Adoption centers on the surrounding compiler and build setup. RAJA requires a C++17 or C++20 compiler, and backend builds rely on an appropriate compiler toolchain because vendor C++ and GPU compilers are not bundled. Teams also operate the library in a self-hosted environment and should align their build configuration with the selected backend. Choose RAJA when portable C++ loop parallelism, reductions, scans and data-layout control matter across CUDA, HIP, OpenMP or SYCL systems. Teams seeking a managed service or a non-C++ SDK should choose a different approach.

RAJA pros and cons

  • Where it wins
    • Separates loop bodies from execution with single-source C++ templates
    • Supports sequential, OpenMP, CUDA, HIP, TBB, OpenMP target and SYCL backends
    • Includes reductions, scans, atomics, sorting, layouts and multidimensional views
  • Where it doesn't
    • Requires a C++17 or C++20 compiler toolchain
    • Does not bundle vendor C++ or GPU compilers
    • Self-hosted deployment requires teams to manage their build environment

RAJA fact sheet, pricing and score →

Advertiser disclosure: iTechGuides is reader-supported. Vendors can pay for top positions in our rankings and for a place on other products' pages, and we may earn a commission when you click some links. Bought places stand at the number bought; this page and our disclosure say so. Paying vendors may get a fuller, more detailed review from our editors, and that review can change their score. How we rank.

Last updated · How we research and update