Computer architecture

About the book

Computer architecture treats design as a set of measured choices. The authors organize the subject around cost, performance and power, then show how those pressures push processor pipelines, caches and interconnects in particular directions. The book's central habit is to compare alternatives with numbers instead of intuition, which gives readers a way to reason about machines they have never seen.

The material spans instruction set design, arithmetic units, memory hierarchy, input and output, and multiprocessors. Parallelism appears as a recurring problem, because the same questions about throughput and latency resurface at every level. Worked examples and quantitative exercises carry much of the argument, so the prose often serves the tables and calculations.

The edition reflects the hardware of its moment. Its framing of reduced instruction sets, cache behavior and shared-memory multiprocessors shaped later teaching, but specific products and benchmark figures have aged. Read it for the reasoning method and the vocabulary, then check current treatments for present-day processor families.

How to read it

Approach the book with a pencil and a calculator. Many sections hinge on cost-performance equations and Amdahl-style reasoning, and the argument becomes clear only when the numbers are worked through. Skim the historical asides on a first pass and return to them once the technical vocabulary is familiar.

The treatments of memory hierarchy and instruction sets reward slow reading, because later material assumes fluency with terms introduced there. The exercises are where the book tests understanding, so treat them as part of the text. Readers without some digital logic and assembly background should expect to pause and fill gaps.

Who it suits

This suits computer science students, working engineers moving toward hardware or performance work, and systems programmers who want to understand why machines behave as they do. Instructors will find a coherent framework for a first architecture course. Readers who prefer conceptual overviews without mathematics will find the density frustrating.

Reading considerations

The edition is dated. Processor examples, benchmark data and discussions of specific machines belong to the late 1980s, so use it alongside a current text for contemporary designs. The quantitative emphasis can obscure qualitative engineering judgment, and some conclusions about instruction set style have been debated since. Check which edition you hold before relying on any figure.