2011 HN Comment Predicted Memory Bandwidth, Not FLOPS, Was the Real Bottleneck for Strong AI
Darpinian · x · 2026-10-01
Darpinian dug up a 2011 Hacker News thread from just before AlexNet, where user modeless argued strong AI is fundamentally a hardware problem — and that memory bandwidth, not FLOPS or storage, is the right benchmark. He estimated the fastest supercomputer of the day would run brain-like algorithms a billion to a trillion times slower than real time, too slow to iterate on the right algorithms at all. A strikingly prescient take on compute bottlenecks before deep learning took off.
More from AGI Musings
- Fei-Fei Li: Those fearing the intelligence explosion haven't considered the stupidity explosion — NaveenGRao · 2026-10-01
- Banning open weight models won't stop attackers, only enterprises — and that's the point — james_mtc · 2026-10-01
- Northwestern launches Economics of AI seminar series, Kleinberg kicks off Oct 6 — soumitrashukla9 · 2026-10-01
- McAfee: EU Tech Regulation Has Left European Productivity 20% Behind the US — amcafee · 2026-10-01
- Pedro Domingos: There are no software engineers anymore, we're all agent wranglers — pmddomingos · 2026-10-01
- Robin Hanson says AI will stay 'normal tech' with no foom; critics push back — pwlot · 2026-10-01