Computer Architecture - Lecture 6: Memory-Centric Computing III (Fall 2025)

Item: Computer Architecture - Lecture 6: Memory-Centric Computing III (Fall 2025)
Rating: 5
Author: PostSingularity Institute

by Geraldo (postdoc, SAFARI Research Group)

Formal & Physical Sciences Computing & Cybersecurity UYComputer science UYFComputer architecture and logic design

🎙 Geraldo (postdoc, SAFARI Research Group) 👥 64K 📅 October 11, 2025 ⏱ 153 min 👁 1K 📄 lecture 🧭 2026-08-16

Available in: English (current) Français

Keywords

processing-in-memorydata movementroofline modelLLC missesmemory-centric computingadoption

Summary

This lecture, part of the Computer Architecture course at ETH Zürich, focuses on the adoption challenges of memory-centric computing, specifically processing-in-memory (PIM). The lecturer, Geraldo, begins by recapping previous lectures on PIM architectures, including processing near memory and processing using memory, and the potential of DRAM-based operations like RowClone and Ambit. The core of the lecture addresses why PIM systems are not yet widespread, highlighting both business and engineering challenges. The main focus is on the engineering obstacles, starting with the need for a holistic system view and the difficulty of identifying suitable applications. The lecture critiques simple metrics like the roofline model and LLC misses per kilo instructions for predicting PIM suitability, showing through experimental data that these metrics are insufficient. A more comprehensive methodology is proposed, combining profiling, memory trace analysis, and scalability analysis to classify applications into six categories of data movement bottlenecks. The lecture emphasizes the need for a shift in mindset from processor-centric to memory-centric computing and discusses the importance of programmability, system integration, and runtime support for PIM adoption. The talk concludes by outlining the steps required to enable PIM at scale, including application characterization, programming models, and system software support.

197 words

Critical Evaluation

Value of the Information & Strength of the Argument

The lecture provides valuable insights into the practical challenges of adopting processing-in-memory, moving beyond theoretical benefits to real-world engineering issues. The argumentation is solid, supported by experimental data from profiling over 100 applications. The critique of existing metrics like the roofline model and LLC misses is well-founded, demonstrating their limitations with concrete examples. The proposed methodology for classifying data movement bottlenecks is a significant contribution, offering a more nuanced approach. The lecture successfully argues that a holistic view and a shift in mindset are necessary for PIM adoption.

97 words

Title / Content Match

Title accurately reflects the content: a lecture on memory-centric computing, specifically focusing on adoption challenges.

Quality & Reliability

9/10

Lecture by a postdoc from a leading research group, based on peer-reviewed publications and extensive experimental data. High technical accuracy and clear methodology.

Key Moments

Markers derived by PSI from the transcript: the creator did not define chapters.

Cited Sources

A Modern Primer on Processing in Memory — Recommended reading for the lecture, providing a comprehensive overview of PIM.
Memory-Centric Computing: Solving Computing's Memory Problem — Recommended reading, discussing the memory problem and memory-centric solutions.
Memory-Centric Computing: Recent Advances in Processing-in-DRAM — Recommended reading, focusing on recent advances in processing-in-DRAM.
Intelligent Architectures for Intelligent Computing Systems — Recommended reading, discussing intelligent architectures for computing systems.
RowHammer: A Retrospective — Recommended reading, providing a retrospective on RowHammer.
Fundamentally Understanding and Solving RowHammer — Recommended reading, discussing fundamental understanding and solutions for RowHammer.
Accelerating Genome Analysis via Algorithm-Architecture Co-Design — Recommended reading, discussing acceleration of genome analysis.
From Molecules to Genomic Variations: Accelerating Genome Analysis via Intelligent Algorithms and Architectures — Recommended reading, discussing intelligent genome analysis.

Concurring Sources

A Modern Primer on Processing in Memory — Provides a comprehensive overview of PIM, aligning with the lecture's content.
Memory-Centric Computing: Solving Computing's Memory Problem — Discusses the memory problem and memory-centric solutions, consistent with the lecture's theme.

External References

Contribution & Novelties

The lecture provides a critical analysis of existing metrics for identifying PIM-suitable applications, showing their inadequacy and proposing a more comprehensive methodology. It emphasizes the need for a holistic system view and a shift in mindset from processor-centric to memory-centric computing. The lecture also highlights the importance of programmability and system integration for PIM adoption.

Pour aller plus loin :

Processing-in-memory (Wikipedia) — Overview of PIM concepts and history.
Roofline model (Wikipedia) — Explanation of the roofline model and its use in performance analysis.
Memory hierarchy (Wikipedia) — Background on memory hierarchy and its impact on performance.

97 words

Radar Profile

The radar profile shows high scores across all dimensions, indicating a well-rounded and rigorous lecture. The high scores in quantity and quality of information reflect the depth and accuracy of the content. The technical level is high, suitable for an advanced audience. The overall reliability is strong, supported by authoritative sources.

Reliability 9/10