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

Kokkos review

Free#25 of 39 in GPU Computing SDKs

A portable C++ layer for parallel CPU and accelerator programming across major GPU backends.

6.8/10Editor score
Kokkos6.8 Visit Kokkos

Reviewed by iTechGuides Editors · Editorial team · Updated Nov 2026

Kokkos is an open-source C++ programming model and library for performance-portable high-performance computing. It abstracts parallel execution and data management so applications can target CPUs, GPUs and other accelerators through interchangeable backends. The project is suited to mid-market and enterprise teams building portable compute software, with support for Linux, Windows and macOS alongside self-hosted deployment.

Its open-source model includes a free plan, and teams can build it from source or through package managers. The ecosystem spans CUDA, HIP, SYCL and OpenACC device backends, plus OpenMP, C++ Threads, HPX and Serial host backends. Kokkos also supplies execution-space and memory-space models, multidimensional Kokkos::View arrays, hierarchical parallelism, and parallel_for, parallel_reduce and parallel_scan dispatch. Native GPU graph API mappings extend its fit for applications that need backend-specific graph integration while retaining a common programming model.

The trade-off is operational and language scope. Kokkos is deployed self-hosted and supports C++, so teams need a suitable compiler toolchain and responsibility for integrating builds into their environments. Documentation and community channels are the stated support routes rather than a commercial support structure. Choose Kokkos when portability across NVIDIA, AMD and Intel architectures matters and your developers can manage C++ infrastructure; consider another option if you need a managed service, commercial support coverage or bindings for languages beyond C++.

Kokkos pros and cons

  • Where it wins
    • Targets CUDA, HIP, SYCL and OpenACC device backends
    • Provides Kokkos::View arrays and parallel_for, reduce and scan dispatch
    • Maps across NVIDIA, AMD and Intel GPU architectures
  • Where it doesn't
    • C++ is the sole supported language
    • Self-hosted deployment puts build and operations on your team
    • Support is limited to documentation and community channels

Kokkos 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