CUDA and Kokkos: Low- and High-Level GPU Programming Models

CUDA is a low-level programming model that requires writing detailed code in C++ or Fortran for NVIDIA GPUs only. Kokkos is a high-level programming model in C++ for writing performance portable applications for all major HPC platforms (e.g., NVIDIA GPUs, AMD GPUs and CPUs, Intel CPUs, ARM CPUs). This workshop will introduce these two approaches and provide a discussion the pros and cons of each.

Workshop format: Presentation with some hands-on exercises

Target audience: Researchers looking to accelerate their C++ or Fortran code using GPUs or multicore CPUs.

Knowledge prerequisites: Some experience with C++ programming would be beneficial.

Hardware/software prerequisites: Participants must have an account on our Adroit cluster to work through the hands-on exercises using Kokkos. Details on requesting an account are found in this guide.

Learning objectives: Participants will develop a sense of which GPU programming model to use for a given problem.

Is your research group looking to accelerate a code using GPUs? Consider applying to the 2025 Princeton Open Hackathon by March 12.

See the full PICSciE/RC spring training program or subscribe to the PICSciE/RC mailing list.

PICSciE/Research Computing | View More Events
Co-hosted with: GradFUTURES

Contact the organizers