Stanford Seminar - New Golden Age for Computer Architecture - John Hennessy

Explore the evolution and future of computer architecture in this Stanford seminar featuring John Hennessy. Trace the journey from early IBM compatibility issues to the rise of RISC architectures and the challenges faced by CISC designs. Examine the impact of Moore's Law, power constraints, and security vulnerabilities on processor development. Delve into the emerging field of Domain Specific Architectures (DSAs) and their potential to revolutionize computing efficiency. Learn about the development of specialized hardware like Google's Tensor Processing Unit (TPU) for machine learning applications. Gain insights into the current state of computer architecture and the promising opportunities that lie ahead in this comprehensive overview of the field's past, present, and future.

Introduction.
Outline.
IBM Compatibility Problem in Early 1960s By early 1960's, IBM had 4 incompatible lines of computers!.
Microprogramming in IBM 360 Model.
IC Technology, Microcode, and CISC.
Microprocessor Evolution • Rapid progress in 1970s, fueled by advances in MOS technology, imitated minicomputers and mainframe ISAS Microprocessor Wers' compete by adding instructions (easy for microcode). justified given assembly language programming • Intel APX 432: Most ambitious 1970s micro, started in 1975.
Analyzing Microcoded Machines 1980s.
From CISC to RISC . Use RAM for instruction cache of user-visible instructions.
Berkeley & Stanford RISC Chips.
"Iron Law" of Processor Performance: How RISC can win.
CISC vs. RISC Today.
From RISC to Intel/HP Itanium, EPIC IA-64.
VLIW Issues and an "EPIC Failure".
Fundamental Changes in Technology.
End of Growth of Single Program Speed?.
Moore's Law Slowdown in Intel Processors.
Technology & Power: Dennard Scaling.
Sorry State of Security.
Example of Current State of the Art: x86 . 40+ years of interfaces leading to attack vectors · e.g., Intel Management Engine (ME) processor . Runs firmware management system more privileged than system SW.
What Opportunities Left?.
What's the opportunity? Matrix Multiply: relative speedup to a Python version (18 core Intel).
Domain Specific Architectures (DSAs) • Achieve higher efficiency by tailoring the architecture to characteristics of the domain • Not one application, but a domain of applications.
Why DSAs Can Win (no magic) Tailor the Architecture to the Domain • More effective parallelism for a specific domain.
Domain Specific Languages.
Deep learning is causing a machine learning revolution.
Tensor Processing Unit v1.
TPU: High-level Chip Architecture.
Perf/Watt TPU vs CPU & GPU.
Concluding Remarks.