Modern GPU Architecture Second Edition: Volume Two Compute Acceleration, Tensor Cores, and Advanced Systems

If you value precision, depth, and professional-grade explanations—you’ll find this book indispensable.

Modern GPU Architecture Second Edition — Volume Two

Compute Acceleration, Tensor Cores, and Advanced Systems

Modern GPUs have evolved far beyond graphics. They are now the engines that power artificial intelligence, deep learning, and large-scale scientific computing. This volume reveals how that transformation happened—and how every part of the GPU is engineered for massive parallel acceleration.

Inside, you’ll explore shader core microarchitecture, warp scheduling, memory hierarchy, tensor and matrix acceleration, systolic arrays, mixed-precision arithmetic, sparsity handling, and multi-GPU interconnects. Each topic is backed by equations, pipeline diagrams, and synthesizable Verilog examples that show how real silicon executes these workloads.

What you’ll gain:

Architectural mastery: Understand how compute units, tensor cores, and memory subsystems coordinate to sustain trillions of operations per second.

Implementation insight: Learn from RTL-level design examples, timing strategies, and scalable datapath structures.

Performance intuition: Analyze bottlenecks, throughput, and scheduling behavior using real hardware principles.

System-level perspective: See how modern GPUs manage coherence, interconnects, and distributed workloads across multi-die architectures.

If Volume One showed you how pixels are rendered, Volume Two shows you how intelligence is computed—how arithmetic becomes architecture and architecture becomes acceleration.

Dense, detailed, and unapologetically precise, this is a book for engineers, researchers, and developers who want to understand GPUs not as black boxes, but as the living machinery of computation itself.