Free course built from Cornell's GPU architecture workshop now shared publicly
idanbeck · x · 2026-10-01
Based on Cornell Virtual Workshop's "Understanding GPU Architecture," the author built a course for a teammate and shared it. Topics: GPU vs CPU design, SIMT and warps, kernels and SMs, GPU memory, with examples spanning Tesla V100, RTX 5000, and Blackwell B200, plus exercises and a quiz.
More from Infra
- Arduino argues sub-$900 embedded boards beat Mac minis for Physical AI agent economics — CatAstro_Piyush · 2026-10-01
- SemiAnalysis injects failures to rate GPU cluster renters — providers differ sharply on recovery — AccBalanced · 2026-10-01
- Redditor runs unattended DeepSeek loops for days: 237M tokens for just $3.48 — dogfoodarchitect · 2026-10-01
- Qwen Flash Next MTP work resumes with official GGUF quants and llama.cpp PR — jacek2023 · 2026-10-01
- Undocumented Strata tip: set default sampling params via a sampling block in run config — KissMyShinyArse · 2026-10-01
- Meta's Loop Scaling Laws: Sparsity Gives ~3x Active-Param Efficiency, Recurrence ~2x on Reasoning — facebook · 2026-10-01