Skip to main content

Ctrl+K

From Cray-1 to Frontier: A working engineer's history of supercomputer architecture

Part I — What 'supercomputer' meant

Week 1 — Before the Word: CDC 6600 and the Idea of “Fast”
Week 2 — The Cray-1 (1976): Vectors Become a Product
Week 3 — Vector Fortran and the Birth of Vectorization

Part II — The vector era

Week 4 — X-MP, Cray-2, Y-MP: When Vectors Met Multiprocessing
Week 5 — Japan: NEC SX, Fujitsu VP, Hitachi S-Series
Week 6 — The Vector Wall: ETA-10, Cray-3, and the Death of Bespoke

Part III — The parallel revolution

Week 7 — The Connection Machine: SIMD Massively Parallel
Week 8 — MIMD MPP: The Killer Micros, Connected
Week 9 — MPI: The Standard That Stuck

Part IV — Commodity and cluster

Week 10 — Beowulf and the Death of Bespoke
Week 11 — ASCI, ASC, and BlueGene: When Power Became the Constraint
Week 12 — The Earth Simulator: When Vectors Briefly Won Again

Part V — Modern heterogeneous

Week 13 — CUDA and SIMT: Vectors Reborn as Threads
Week 14 — Exascale: Frontier, Aurora, El Capitan, Fugaku
Week 15 — Anatomy of a 2025 Supercomputer

Capstone

Week 16 — Capstone: SAXPY Across the Eras

Appendix

Appendix A — The Modern Supercomputer vs. the Frontier-LLM Training Cluster

Reference

References
Assessment and Project Guide

Search

Ctrl+K

© Copyright 2026 Duane Moore.