https://www.harshinimaddala.com/writing/ Notes on hardware, AI systems, and other things I’m trying to understand. en https://www.harshinimaddala.com/writing/why-more-flops-arent-enough/ https://www.harshinimaddala.com/writing/why-more-flops-arent-enough/ Why peak compute tells only part of the story, and what memory, data movement and workload design have to do with making LLMs faster. Sat, 24 Oct 2026 09:38:00 GMT