NVIDIA HPC SDK

Type: Platform Tags: NVIDIA, HPC SDK, compilers, CUDA, Fortran, OpenACC, OpenMP, stdpar, math libraries, HPC Related: NVIDIA-HPC-Compilers, CUDA-Fortran, NVIDIA-Fortran-CUDA-Interfaces, NVIDIA-OpenACC, NVIDIA-Stdpar, NVIDIA-CUDA, NVCC, NVPL, cuFFTMp, cuTENSORMg, cuTENSORMp, NVIDIA-HPC-X, NCCL, NVSHMEM, NVTX, CUDA-GDB, Nsight-Compute, Nsight-Systems, Compute-Sanitizer, NGC Sources: https://docs.nvidia.com/hpc-sdk/index.html, https://docs.nvidia.com/hpc-sdk/compilers/hpc-compilers-user-guide/index.html Last Updated: 2026-04-29

Summary

NVIDIA HPC SDK is NVIDIA’s current software development kit for accelerated HPC applications. The 26.3 documentation hub organizes HPC compilers, programming models, math libraries, communication libraries, debugging/profiling tools, container guidance, and CUDA documentation into one SDK surface for scientists and engineers building on NVIDIA GPUs and Arm/x86 server CPUs.

Detail

Purpose

HPC codes often combine Fortran, C, C++, MPI, OpenACC/OpenMP directives, CUDA, math libraries, profiling tools, and deployment scripts. NVIDIA HPC SDK packages those pieces into one supported developer stack so teams can port, optimize, profile, and deploy CPU/GPU HPC applications without assembling separate compiler and library ecosystems manually.

Current scope

NVIDIA context

HPC SDK is the bridge between low-level CUDA and large scientific/HPC application development. It keeps CUDA and CUDA-X libraries close to NVIDIA compilers, directive programming, Fortran support, MPI/communication stacks, and profiler/debugger tooling.

Connections

Resources