cudatoolkit Overview

The CUDA Toolkit is a C and C++ language development environment for CUDA-enabled GPUs. The CUDA Toolkit provides access to many tools including the nvcc compiler, debuggers, profilers, and GPU accelerated libraries. Refer to the Nvidia website for more information about CUDA, or see examples of CUDA and OpenACC accelerated code on the OLCF tutorials page.


To use the cudatoolkit, run
$ module load cudatoolkit
To load the CUDA module. This will give provide access to all the cudatoolkit's tools including the nvcc compiler,  the cuda-gdb debugger, the Nvidia profiler (nvprof), and many other tools. To compile your CUDA code, use nvcc:
$ nvcc -o a.out
Running the compiled Cuda binary on Titan is the same as submitting a normal job since each node contains a gpu.  But on Rhea, extra steps need to be taken to target the nodes that contain GPUs. Refer to the Rhea user guide for details


  • 7.5.18
  • 8.0.44


  • 7.0.28-1.0502.10280.4.1
  • 7.5.18-1.0502.10743.2.1