Python and NumPy have significantly lowered the barrier for entry for developing complex scientific applications. These applications are often limited to executing on a single CPU which can quickly become a performance bottleneck and limit the size of datasets they can work on. We give an introduction to cuNumeric, a drop-in replacement for NumPy which allows applications to scale to multiple nodes and accelerates them using GPUs.
Back to Workshop III: Complex Scientific Workflows at Extreme Computational Scales