NVIDIA cuPyNumeric brings zero-code-change scaling to multi-GPU and multi-node (MGMN) accelerated computing. It is designed to be a drop-in replacement library for NumPy, enabling distributed and accelerated computing on the NVIDIA platform for the Python community. Researchers and scientists can write their programs using native Python and familiar tools without worrying about parallel or distributed computing. cuPyNumeric, layered on top of Legate, allows seamless scaling from single-CPU computers to MGMN supercomputers without code modifications.

Features

• Native Python and NumPy Interface: Supports native Python language and NumPy interface without constraints.
• Automatic Parallelism and Acceleration: Provides automatic parallelism and acceleration for multiple nodes across CPUs and GPUs.
• Seamless Scaling: Scales from one CPU up to thousands of GPUs optimally.
• Legate Integration: Layers on top of Legate, which runs on the CUDA runtime system, providing scalable implementations of popular domain-specific APIs.
• Real-Time Processing: Enables real-time processing of large datasets, such as multi-view lattice light-sheet microscopy data.
• Performance: Offers significant performance improvements, such as up to 2500X speedups for large-scale simulations on GPUs.

Benefits

• Productivity: Allows researchers to write programs productively using native Python without worrying about parallel or distributed computing.
• Scalability: Easily scales programs from single-CPU computers to MGMN supercomputers without code changes.
• Efficiency: Improves operational efficiencies with automatic parallelism and acceleration.
• Flexibility: Suitable for a wide range of applications, from data science and machine learning to computational fluid dynamics.
• Accessibility: Democratizes computing by making it possible for all programmers to leverage the power of large clusters of CPUs and GPUs.