nvhpc-nompi

Version:

23.1

Category:

compiler

Cluster:

Loki

Author / Distributor

https://developer.nvidia.com/hpc-sdk

Description

The NVIDIA HPC SDK (NVHPC) 23.1 is a comprehensive suite of compilers, tools, and GPU-accelerated libraries for developing high-performance applications on NVIDIA hardware.

This variant, `nvhpc-nompi`, includes the full NVHPC toolchain without an MPI stack. It is ideal for:

  • Standalone GPU-accelerated development using CUDA, OpenACC, and OpenMP

  • Users who want to use external/custom MPI implementations (e.g., OpenMPI, MPICH, HPCX)

  • Lightweight compilation workflows without bundled MPI libraries

Key features:

  • Compilers: nvc, nvc++, nvfortran

  • CUDA 12.0 support (with Hopper compatibility)

  • OpenACC 2.7, OpenMP 4.5/5.x target offloading

  • GPU math libraries: cuBLAS, cuFFT, cuSOLVER, cuSPARSE

  • Nsight Compute and Nsight Systems tools (CLI-accessible)

Documentation

$ nvc --help         # For C programs
$ nvc++ --help       # For C++ programs
$ nvfortran --help   # For Fortran programs

NVIDIA HPC SDK documentation:
https://docs.nvidia.com/hpc-sdk/

To verify GPU access:
$ nvaccelinfo

Examples/Usage

  • Load the module:

$ module load nvhpc-nompi/23.1

  • Compile a GPU-enabled C++ program:

$ nvc++ -cuda -O3 main.cu -o main_gpu

  • Compile with OpenACC:

$ nvc -acc -Minfo=acc compute.c -o compute_acc

  • Compile Fortran with OpenMP GPU offloading:

$ nvfortran -mp=gpu solver.f90 -o solver

  • Unload the module:

$ module unload nvhpc-nompi/23.1

Installation

he `nvhpc-nompi/23.1` module was installed from the NVIDIA HPC SDK 23.1 archive: https://developer.nvidia.com/nvidia-hpc-sdk-downloads