Comp. Arch. - Lecture 29: SIMD and GPU Architectures (Fall 2025)
Keywords
Summary
131 words
Critical Evaluation
Value of the Information & Strength of the Argument
The lecture provides high-value information, systematically explaining SIMD and GPU architectures from fundamental concepts to modern implementations. The argumentation is solid, building on established principles and illustrating them with concrete examples like the TPU and GPU designs. The lecturer effectively contrasts different paradigms and highlights trade-offs, such as the cost-efficiency of vector processors versus the speed of array processors. The discussion of time-space duality is particularly insightful, clarifying the design choices in SIMD implementations. The lecture also connects to broader research themes, such as memory-centric computing, reinforcing the importance of the material.
101 words
Title / Content Match
The title accurately reflects the content, covering SIMD and GPU architectures in depth.
Quality & Reliability
9/10
Lecture by a leading academic in computer architecture, based on established concepts and seminal papers, with references to verifiable sources and course materials.
Key Moments
Markers derived by PSI from the transcript: the creator did not define chapters.
- Introduction to SIMD and GPU architectures, overview of lecture topics.
- Discussion of systolic arrays and Google TPU as a real-world example.
- Introduction to Flynn's taxonomy and classification of computer architectures.
- Explanation of SIMD paradigm and data-level parallelism.
- Comparison of array processors and vector processors, time-space duality.
- Detailed discussion of vector registers and vector instruction execution.
- Transition to GPU architectures, overview of GPU organization.
- GPU programming model, warps, and SIMT execution.
- Memory hierarchy in GPUs, importance of bandwidth.
- Recent trends, processing-in-memory, and memory-centric computing.
Cited Sources
- A Modern Primer on Processing in Memory — Recommended reading on processing-in-memory, relevant to memory-centric computing.
- Memory-Centric Computing: Solving Computing's Memory Problem — Recommended reading on memory-centric computing.
- Memory-Centric Computing: Recent Advances in Processing-in-DRAM — Recommended reading on recent advances in processing-in-DRAM.
- Intelligent Architectures for Intelligent Computing Systems — Recommended reading on intelligent architectures.
- RowHammer: A Retrospective — Recommended reading on RowHammer, a memory reliability issue.
- Fundamentally Understanding and Solving RowHammer — Recommended reading on RowHammer.
- Accelerating Genome Analysis via Algorithm-Architecture Co-Design — Recommended reading on accelerating genome analysis.
- From Molecules to Genomic Variations: Accelerating Genome Analysis via Intelligent Algorithms and Architectures — Recommended reading on genome analysis acceleration.
- Course Schedule — Course page for Computer Architecture at ETH Zürich.
- Lecture 29a Slides (PDF) — Slides for the SIMD part of the lecture.
- Lecture 29a Slides (PPTX) — Slides for the SIMD part of the lecture.
- Lecture 29b Slides (PDF) — Slides for the GPU part of the lecture.
- Lecture 29b Slides (PPTX) — Slides for the GPU part of the lecture.
Concurring Sources
- A Modern Primer on Processing in Memory — Supports the discussion on memory-centric computing.
- Memory-Centric Computing: Solving Computing's Memory Problem — Supports the discussion on memory-centric computing.
- Memory-Centric Computing: Recent Advances in Processing-in-DRAM — Supports the discussion on processing-in-DRAM.
External References
Contribution & Novelties
The lecture provides a thorough and structured introduction to SIMD and GPU architectures, synthesizing fundamental concepts with modern implementations. It offers a clear comparison of array and vector processors, emphasizing the time-space duality, and explains how GPUs combine both approaches. The discussion of memory-centric computing and processing-in-memory highlights current research directions. The lecture is an excellent educational resource for understanding data-level parallelism and its role in modern computing.
Pour aller plus loin :
- Flynn’s taxonomy — Classification of computer architectures.
- SIMD — Overview of single instruction, multiple data.
- Graphics processing unit — General information on GPUs.
- Vector processor — Explanation of vector processing.
- Processing-in-memory — Concept of integrating computation with memory.
111 words
Radar Profile
The radar profile shows high scores across all dimensions, indicating a comprehensive and reliable lecture. The balance between quantity and quality of information, technical depth, and overall reliability is consistent, making it an excellent resource for advanced learners.