Reproducing classic AI models now costs pocket change: AlexNet ~$2, GPT-2 ~$30
cneuralnetwork · x · 2026-11-11
Training AI models once required expensive hardware and days of compute — AlexNet took 5–6 days on 2 GTX 580s, BERT-base 4 days on 16 TPU chips, GPT-2 small trained on 40GB of WebText. Today reproducing them costs roughly $2, $118, and $30 respectively, per a cost breakdown chart shared by @cneuralnetwork, illustrating how cheap compute has made landmark models reproducible by anyone.
More from Infra
- Natural language search over autonomous driving data with Cosmos-Embed1 + OpenSearch — 4310sy · 2026-11-11
- Fully local MiniMax H3 Image + LTX-2.5 pipeline generates a shot in under 5 minutes on one RTX 5090 — Hailuo_AI · 2026-11-11
- Dev claims ~700 tokens/s running Qwen 3 27B on a single RTX 5090 — IgorCarron · 2026-11-11
- TSMC Co-COO Kevin Zhang: demand for leading-node chips is 'insatiable' — firstadopter · 2026-11-11
- Sable proves bit-exact LLM inference: Qwen2.5-32B verified on one GPU in 92.5s — MoonMath-ai · 2026-11-11
- Investor makes the case: HBM at 3x forward earnings with 60% growth is undervalued — firstadopter · 2026-11-11