Turing Award Winner: TPUs vs GPUs vs CPUs, Computer Architecture, RISC vs CISC | David Patterson

2026-07-06 · Source: The Peterman Post · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering, Cloud Computing & IT Infrastructure · Depth: Advanced, extended

Summary

Turing Award winner David Patterson provides a comprehensive overview of computer architecture evolution, contrasting RISC and CISC designs. He explains that RISC ultimately "won" the architectural debate, with ARM processors now dominating mobile devices and increasingly entering the cloud, accounting for 99% of all processors today. Patterson details the rise of domain-specific architectures like GPUs, initially for graphics, and later Google's Tensor Processing Units (TPUs) in 2016, which were purpose-built for machine learning's matrix multiplication. TPUs achieved 30 times better inference performance than contemporary GPUs and 80 times better than CPUs by innovating with custom floating-point formats like "brain float 16" and eliminating hardware caches. He also addresses the slowing of Moore's Law, noting that while transistor doubling has ceased, technological improvements continue through specialized components and advanced packaging like chiplets.

Key takeaway

For AI Architects evaluating hardware for machine learning, recognize that general-purpose CPUs are no longer the performance leaders. You should prioritize domain-specific accelerators like TPUs or GPUs, which offer significantly higher performance and energy efficiency for AI workloads. Your strategy should account for the slowing of Moore's Law by focusing on specialized architectures and advanced packaging, rather than expecting continuous general-purpose CPU improvements.

Key insights

Specialized architectures like TPUs and GPUs now drive performance gains, surpassing general-purpose CPUs due to Moore's Law slowdown.

Principles

• RISC designs offer superior speed and energy efficiency over CISC.
• Domain-specific architectures yield dramatic performance for narrow tasks.
• Post-Moore's Law gains stem from specialization and advanced packaging.

Method

Google's TPU design method involved identifying matrix multiplication as the core ML operation, building a dedicated unit, removing general-purpose features like hardware caches, and innovating floating-point formats.

In practice

• Utilize ARM for energy-efficient mobile and cloud deployments.
• Deploy TPUs or GPUs for machine learning to achieve peak performance.
• Consider specialized hardware for critical domain-specific computations.

Topics

• Computer Architecture
• RISC Architecture
• ARM Processors
• Tensor Processing Units
• GPU Acceleration
• Moore's Law

Best for: CTO, VP of Engineering/Data, Director of AI/ML, AI Scientist, Machine Learning Engineer, AI Architect

Related on AIssential

• CPU vs GPU vs TPU — Hardware specialization (CPU, GPU, TPU) dictates optimal performance for different computational workloads.
• ARM vs x86: Everything You Need to Know in 4 Minutes — ARM's RISC design and SoC integration offer superior efficiency and sustained performance compared to x86's CISC, despite x86's software legacy.
• Low-Energy Reduced RISC-V Instruction Subset Processor for Tsetlin Machine Inference at the Edge — A specialized RISC-V processor significantly boosts Tsetlin Machine inference efficiency and programmability for edge AI applications.
• The Hardware That Makes AI Possible — Modern AI's progress is fundamentally enabled by specialized hardware optimized for parallel computation, beyond just algorithmic advances.
• How Google’s Tensor Processing Unit (TPU) Works? — Specialized hardware like TPUs dramatically boosts AI performance by optimizing for specific neural network operations.
• CPU vs GPU — GPUs' parallel matrix and vector processing architecture provides a significant performance advantage for AI workloads over CPUs.

See Counsel's argued verdicts on the open AI decisions leaders are weighing →

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by The Peterman Post.