NVIDIA: Open Source Ecosystem

Much of today's GPU-accelerated computing happens through a growing ecosystem of open source libraries, but figuring out what already exists and how these pieces fit together can be challenging. NVIDIA: Open Source Ecosystem gives data scientists, ML engineers, and systems developers a practical lay of the land, helping them understand and utilize expertly crafted and optimized libraries for their problems. With these libraries, users can benefit from "speed-of-light" computing without the need for writing low-level performance-sensitive CUDA code.

Author Max Buckley connects GPU architecture to the design of tools such as CUDA-X Data Science, CUTLASS, NCCL, TensorRT-LLM, Triton Inference Server, and many others. You'll learn how to accelerate familiar Python workflows, build fast and efficient data pipelines, scale across multiple GPUs, and serve optimized models in production. Along the way, you'll discover when these higher-level libraries are enough, when to go deeper, and how to make informed choices as the ecosystem continues to evolve.

• Move beyond Pandas and Scikit-learn workflows with CUDA-X Data Science
• Build end-to-end pipelines that keep data on the GPU with DALI
• Profile and optimize GPU workloads systematically with Nsights
• Scale workloads across multiple GPUs and nodes using NCCL
• Deploy efficient LLM inference with TensorRT-LLM and the Triton Inference Server
• Contribute to this ecosystem through GitHub
• Choose the right NVIDIA libraries to accelerate your workload