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-HPC-Compilers:
nvc,nvc++,nvfortran, and NVCC documentation. - Programming models: C++ parallel algorithms/stdpar, CUDA C++, CUDA-Fortran, NVIDIA-Fortran-CUDA-Interfaces, NVIDIA-OpenACC, and OpenMP.
- Math libraries: cuBLAS, cuBLASMp, cuTENSOR, cuTENSORMg, cuTENSORMp, cuSPARSE, cuSOLVER, cuSOLVERMp, cuFFT, cuFFTMp, cuRAND, and NVPL.
- Communication libraries: NCCL, NVSHMEM, and NVIDIA-HPC-X MPI.
- Tools: CUDA-GDB, Nsight-Compute, Nsight-Systems, Compute-Sanitizer, and NVTX.
- Containerization through the HPC SDK container guide and NGC.
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
- NVIDIA-HPC-Compilers - compiler family inside the SDK.
- CUDA-Fortran - explicit CUDA programming model for Fortran developers in the SDK.
- NVIDIA-Fortran-CUDA-Interfaces - Fortran modules and interfaces for CUDA libraries.
- NVIDIA-OpenACC - NVIDIA HPC compiler implementation of OpenACC GPU offload.
- NVIDIA-Stdpar - standard-language parallelism path in
nvc++andnvfortran. - NVIDIA-CUDA - core GPU computing platform underneath the SDK.
- NVCC - CUDA C/C++ compiler driver also listed from the HPC SDK docs.
- NVPL - CPU math library collection included in the current SDK documentation hub.
- cuFFTMp - distributed multi-process FFT library distributed through HPC SDK.
- cuTENSORMg and cuTENSORMp - cuTENSOR multi-GPU and distributed tensor contraction paths.
- NCCL, NVSHMEM, and NVIDIA-HPC-X - communication stack for multi-GPU and multi-node HPC jobs.
- CUDA-GDB, Nsight-Compute, Nsight-Systems, and Compute-Sanitizer - development and profiling tools exposed through the SDK.
- NGC - container distribution context for SDK and HPC software.