AI Systems Performance Engineering (Full Colour Edition)

AI Systems Performance Engineering by Chris Fregly looks at how to make AI systems run efficiently across hardware, software and algorithms. It is written for engineers, researchers and developers who work with large generative models. The book gives practical steps for tuning GPU CUDA kernels, PyTorch-based algorithms, and training and inference systems that span many nodes. It also covers scaling GPU clusters, distributed training jobs and inference servers, aiming for systems that are resilient, scalable and cost-effective. A checklist of 175-plus optimizations closes the book.

Why You Should Read?

• Learn to co-optimize hardware, software and algorithms so your AI systems stay efficient in both training and inference.
• Master the scaling of GPU clusters, distributed model training jobs and inference servers for high performance.
• Use the closing checklist of 175-plus proven optimizations as a ready-to-use reference for your own systems.