-
Book Overview & Buying
-
Table Of Contents
Modern Computer Architecture and Organization - Third Edition
By :
This chapter examined the computing architectures that enable advanced artificial intelligence systems, focusing on the design, training, and deployment of LLMs. Using GPT-2 as a representative example, we explored the internal structure of transformer-based models and the computational demands associated with their operation.
We analyzed the distinct requirements of LLM training and inference and discussed how GPU hardware, which is optimized for massively parallel matrix computations, provides a foundation for efficient processing. The chapter described how GPU and TPU clusters are interconnected to form large-scale LLM supercomputers capable of handling the immense workloads of state-of-the-art models.
We reviewed the datacenter infrastructure that powers today's AI applications, including the systems that host and serve popular models such as ChatGPT. These facilities integrate advanced networking, power delivery, and cooling technologies to meet the growing demand for...