GTC 2020: Running Unmodified NumPy Programs on Hundreds of GPUs with Legate NumPy
After clicking “Watch Now” you will be prompted to login or join.
Click “Watch Now” to login or join the NVIDIA Developer Program.
Running Unmodified NumPy Programs on Hundreds of GPUs with Legate NumPy
Mike Bauer, NVIDIA
Learn how you can run unmodified NumPy programs on hundreds of GPUs with Legate NumPy. We'll describe the implementation of Legate NumPy, a drop-in replacement for NumPy that lets you transparently distribute and accelerate NumPy programs across machines ranging from a single DGX box to the world's top supercomputers. To illustrate the power of Legate NumPy, we'll present performance results on a combination of HPC and traditional machine-learning applications written in NumPy running at scale. In addition to describing how to use Legate NumPy, we'll also describe its underlying architecture and how it leverages the Legion runtime system to make distributed and accelerated NumPy possible. Finally, we'll cover the underlying Legate interface on which Legate NumPy is built and how you can construct your own accelerated and distributed Python libraries using Legate to seamlessly compose and interoperate with Legate NumPy.