NVIDIA HPC Compilers
Type: Tool Tags: NVIDIA, HPC SDK, compiler, nvc, nvc++, nvfortran, CUDA, OpenACC, OpenMP, stdpar, Fortran, C, C++ Related: NVIDIA-HPC-SDK, CUDA-Fortran, NVIDIA-Fortran-CUDA-Interfaces, NVIDIA-OpenACC, NVIDIA-Stdpar, NVIDIA-CUDA, NVCC, libNVVM, NVVM-IR, PTX-ISA, CUDA-Unified-Memory, Nsight-Compute, Nsight-Systems Sources: https://docs.nvidia.com/hpc-sdk/compilers/hpc-compilers-user-guide/index.html, https://docs.nvidia.com/hpc-sdk/compilers/hpc-compilers-ref-guide/index.html, https://docs.nvidia.com/hpc-sdk/index.html Last Updated: 2026-04-29
Summary
NVIDIA HPC Compilers are the Fortran, C++, and C compiler family inside NVIDIA-HPC-SDK. The current docs identify NVFORTRAN (nvfortran), NVC++ (nvc++), and NVC (nvc) as compilers for NVIDIA GPUs and x86-64/Arm server CPUs, with support for CUDA-oriented programming, OpenACC, OpenMP, standard parallelism, optimization flags, MPI workflows, and deployment guidance.
Detail
Purpose
Large HPC applications rarely move to GPUs through CUDA C++ alone. The NVIDIA HPC compilers let scientific codes keep Fortran, C, and C++ source structure while using GPU offload through CUDA-Fortran, NVIDIA-OpenACC, OpenMP target directives, or NVIDIA-Stdpar. They also provide the command-line optimization, linking, runtime, and deployment controls needed for production HPC builds.
Current compiler commands
nvfortran: NVIDIA Fortran compiler for NVIDIA GPUs and AMD, Intel, and Arm CPUs; supports ISO Fortran 2003, many Fortran 2008 features, CUDA Fortran, OpenACC, OpenMP, and Fortran standard parallel language features.nvc++: NVIDIA C++17 compiler for NVIDIA GPUs and AMD, Intel, and Arm CPUs; supports C++17 parallel algorithms, OpenACC, and OpenMP.nvc: NVIDIA C11 compiler for NVIDIA GPUs and AMD, Intel, and Arm CPUs; supports OpenACC GPU programming and multicore CPU programming through OpenACC/OpenMP.- NVCC: CUDA C/C++ compiler driver listed alongside the HPC compiler family in the current HPC SDK docs.
Current scope
- Command-line options, filename conventions, data types, platform considerations, site/user rc files, and common build tasks.
- GPU execution model, memory models, compute capability targeting, CUDA toolkit selection, and PTX JIT compilation.
- OpenACC, OpenMP, stdpar C++/Fortran, MPI, library creation/linking, environment variables, and redistribution.
- Reference details for compiler options, language extensions, runtime environment, C++ dialect support, intrinsics, and messages.
Connections
- NVIDIA-HPC-SDK - parent SDK documentation surface.
- CUDA-Fortran - explicit Fortran CUDA programming model compiled by
nvfortran. - NVIDIA-Fortran-CUDA-Interfaces - Fortran modules and interfaces for CUDA library calls.
- NVIDIA-OpenACC - directive-based GPU offload path supported by
nvc,nvc++, andnvfortran. - NVIDIA-Stdpar - C++ and Fortran standard parallelism path in NVIDIA compilers.
- NVCC - CUDA C/C++ compiler driver adjacent to the HPC compiler family.
- libNVVM, NVVM-IR, and PTX-ISA - compiler backend and virtual ISA context for GPU code generation.
- CUDA-Unified-Memory - memory mode used by OpenACC/OpenMP/stdpar guidance.
- Nsight-Compute and Nsight-Systems - profiler/debugger context for compiler-generated GPU code.