Fundamentals of CPU/GPU Power and Performance Modeling

Erscheinungsjahr 2026
ISBN: 978-3-032-34111-2
Verlag: Springer

Buch, Englisch, 1034 Seiten, Format (B × H): 178 mm x 254 mm

This book presents a comprehensive, academically rigorous, and industry-grounded framework for modeling power and performance in modern CPUs and GPUs. It bridges architectural theory, workload-driven performance analysis, and realistic chip-level power modeling, explicitly incorporating variations and constraints that are often treated as secondary or external—such as thermals, yield, reliability, and power delivery. By integrating these dimensions into a single framework, the book connects early architectural decisions to post-silicon behavior and manufacturing out-comes, and demonstrates how accurate power–performance (PnP) modeling can directly inform high-impact decisions including binning strategy, product stacking, and wafer utilization.

Zielgruppe

Professional/practitioner

Autoren/Hrsg.

Elsayed, Rany

Fachgebiete

• Mathematik | Informatik EDV | Informatik Informatik Rechnerarchitektur
• Mathematik | Informatik EDV | Informatik Technische Informatik Eingebettete Systeme
• Technische Wissenschaften Elektronik | Nachrichtentechnik Elektronik Bauelemente, Schaltkreise

Inhaltsverzeichnis

Microprocessor Topology Evolution.- Multi-Core Microprocessor Architecture.- Uncore Architecture.- Workload Classification and Characterization.- CPU/GPU Performance Modeling.- CPU/GPU Power Management.- Power Modeling Fundamentals.- Architecture Power Models.- Reliability Modeling.- SoC Thermal Modeling.- Die Yield Modeling.- Power Delivery Network Design.- Platform Design and Modeling.- SoC Power-Performance Conver-gence and Optimization.- Binsplit and Wafer Utilization Analysis.

Über Autor(innen)

Rany T. Elsayed is a Principal Engineer at NVIDIA, where he works on power, performance, and system-level modeling for advanced AI compute platforms. He previously served as a Principal Engineer and Director in Intel’s Data Center and AI Group, where he led power, thermal, performance, and binsplit modeling for Intel Xeon processors and AI accelerator platforms. With more than twenty years of experience spanning semiconductor architecture, process integration, and system-level modeling, he is widely recognized as Intel’s leading expert in SoC power and performance modeling.

Dr. Elsayed has played a central role in the definition and optimization of six generations of Xeon and AI products, covering the full lifecycle from early architecture pathfinding through manufacturing, SKU stack engineering, and data-center total cost of ownership (TCO) optimization. His work unifies SoC macro-architecture, process technology, power delivery, thermals, workload characterization, variations, yield, and reliability into predictive modeling frameworks that directly guide product definition and platform decisions. He has deep hands-on expertise in binsplit optimization, wafer utilization, power delivery specification, thermal throttling strategies, and reliability-aware performance scaling across server and accelerator products.

Before joining NVIDIA, Dr. Elsayed held senior technical leadership roles across Intel’s Data Center and AI Group, Intel Custom Foundry, Intel’s Nano-Lithography Pathfinding Group, and Intel’s Advanced Design Group. Across these roles, he led research and development in design–technology co-optimization (DTCO), advanced process characterization, lithography modeling, interconnect modeling, and full-chip power, performance, and yield analysis across multiple technology nodes. His career reflects a rare breadth, bridging first-principles physics-based modeling, circuit and device design, and system-level performance analysis.

Dr. Elsayed received his Ph.D. in Electrical and Computer Engineering from Carnegie Mellon University and holds graduate and undergraduate degrees from Cairo University. He is a Senior Member of IEEE, author of more than twenty refereed technical publications, and inventor on multiple U.S. patents. He has received numerous divisional and corporate awards for technical excellence. Fundamentals of CPU/GPU Power and Performance Modeling distills more than two decades of industrial and academic experience into a unified, practitioner-oriented framework for modeling and optimizing modern compute systems.