Episode Details

Episode 3 of our 10-part series on building an AI computer from scratch. Today we tackle the strangest thing in the whole build: graphics cards are the artificial intelligence component. This seems backwards until you realize graphics rendering and neural networks solve identical problems—millions of trivial, independent calculations that happen simultaneously. The GPU wasn't repurposed for AI. It was always built for this shape of problem. 00:00:00 - Intro: Three of ten episodes in the build 00:02:30 - The bizarre core fact: graphics cards run AI 00:06:15 - Why this seems backwards 00:08:45 - The revelation: pixels and neurons are structurally identical 00:12:30 - How parallel processing works 00:16:00 - Why nobody repurposed anything --- Sources & further reading: • ComputerBuild repo — decisions/decision-log.md, D1 (NVIDIA/CUDA only; the 2024 ROCm failure; reversal condition) • ComputerBuild repo — workloads/workload-map.md, section A (Kaggle-offloaded workloads and their measured VRAM ceilings) • ComputerBuild repo — inventory/machines.md (laptop specs; the Radeon iGPU as origin of D1) • ComputerBuild repo — docs/build-recommendation.md, section 7 (thermals, noise and idle power as unresearched) This podcast episode was fully generated by AI — research, script, voices, and production. Built with Claude, Piper TTS, and automated pipeline tooling.