Porting Scientific Research Codes to GPUs with CUDA Fortran

Print Friendly, PDF & Email

Josh Romero from NVIDIA

In this video from the Stanford HPC Conference, Josh Romero from NVIDIA presents: Porting Scientific Research Codes to GPUs with CUDA Fortran: Incompressible Fluid Dynamics Using the Immersed Boundary Method.

“In this session, we intend to provide guidance and techniques for porting scientific research codes to NVIDIA GPUs using CUDA Fortran. The GPU porting effort of an incompressible fluid dynamics solver using the immersed boundary method will be described. Several examples from this program will be used to illustrate available features in CUDA Fortran, from simple directive-based programming using CUF kernels to lower level programming using CUDA kernels. In addition to this, an overview of available profiling and optimization tools will be provided. The session will conclude with the results of several test cases using the new GPU-based solver, with performance comparisons to the original CPU-based implementation.”

Josh Romero is a Developer Technology Engineer at NVIDIA in the HPC Software and Benchmarks group. He recently received his PhD in Aerospace Engineering from Stanford University with doctoral research focused on high-order finite element methods for computational fluid dynamics (CFD) and their application to GPUs.

Josh Romero will also be a featured presenter at the GPU Technology Conference in San Jose.

See more talks in the Stanford HPC Conference Video Gallery

Check out our insideHPC Events Calendar


  1. Great video, thanks for sharing. I’ve discovered for myself the trick of transposing 3d CFD datasets while working with Xeon Phi’s (‘only’ 224 threads). I wasn’t sure how much it helps in cuda Fortran.