Simplify Multi-GPU Programming with Task-Based Runtimes | NVIDIA GTC 2025

Video Description

Learn how to scale your application across multiple GPUs without writing message-passing code. This GTC 2025 session introduces task-based runtime systems that let you write multi-GPU libraries more easily and scale transparently—no explicit communication or coordination required. Plus, see how to integrate your code with domain libraries in the same program. Speaker: Manolis Papadakis, Software Engineering Manager, Legate Framework, NVIDIA CUDA Toolkit: https://developer.nvidia.com/cuda-toolkit Watch more NVIDIA GTC sessions on demand: https://www.nvidia.com/en-us/on-demand/?ncid=so-yout-194474-vt33 Replay of NVIDIA GTC 2025 session: S72579 Topic: Models / Libraries / Frameworks - Accelerated Computing Libraries NVIDIA technology: #CUDA, #CUDA-X #cuda #gtc25 #nvidia #gtc