Using CuPyNumeric and the Legate Ecosystem for Multi-GPU Scaling on Marlowe

Wednesday, August 13, 2025 | 2:00 - 3:30 PM PDT CoDa Building, Rm E401, 4th Floor

Abstract

Discover how to use CuPyNumeric for seamlessly scaling NumPy code to multi-GPU and multi-node setups. This workshop will discuss the best practices and uses of cuPyNumeric with real world examples and performance comparisons.

Speakers

Marcus Manos is a Solutions Architect at NVIDIA working with Financial Services customers. He received his master’s in Data Science from the University of California, Berkeley, where he trained deep learning models. While at NVIDIA, Marcus has worked on a diverse set of problems involving LLM inference, reasoning model training, and cuPyNumeric.

Manolis Papadakis is a Software Engineering Manager at NVIDIA. Manolis joined NVIDIA after completing his Ph.D. in computer science at Stanford University, with a focus on distributed and scientific computing. He has since been working on making Legate/cuPyNumeric a reality.

Part of the ongoing NVIDIA & Marlowe seminar series.