Dev builds Cray-2 emulator from scratch, runs a GPT-3-class LLM at 1 token per 4s
inductionheads · x · 2026-11-08
Developer InverseMarcus found no runnable emulator of the 1985 Cray-2 supercomputer existed, so he wrote one from Cray's official hardware manual — an instruction-level emulator of the four-processor system with a clock-period timing model and a CAL Version 2 assembler.
He then implemented the modern open-weight model Ternary-Bonsai-8B on the emulator. Based on cycle-time calculations, it generates roughly 1 token every 4 seconds, with intelligence approximately equal to GPT-3 (stronger reasoning, weaker trivia recall due to model size). His takeaway: with the right weights, 1985 could have had proto-AGI. Both the emulator and LLM implementation are open-sourced on GitHub under MIT license.
More from Fun
- The AI Papers Iceberg: From Attention to Obscure RL Papers in 7 Tiers — burny_tech · 2026-11-08
- User turns ChatGPT dot into a 24/7 task slave: 'figure it out, don't ask me questions' — filiksyos · 2026-11-08
- What every SF coffee shop looks like: hippos pitching AI agents to VCs — dunkhippo33 · 2026-11-08
- Nvidia's 2006 GPU fluid demo now runs on a 2022 phone as a 3D voxel sim — pvncher · 2026-11-08
- AI Model Fan Art Chart Goes Viral: DeepSeek, ChatGPT, Claude, Gemini, Grok as Anime Girls — Boring_Aioli7916 · 2026-11-08
- "Serverina" AI Video Delights Reddit: Maltese Dog Working as a Restaurant Server — TradingCardGirl · 2026-11-08