Archive — US — turtlesai
An LLM on Windows 98: AI Returns to Its Origins
Researchers successfully ran an advanced LLM on a Windows 98 PC with a Pentium II processor and 128 MB RAM, achieving 39.31 tokens per second, thanks to BitNet architecture. This project aims to democratize AI access and encourages development on older devices, showcasing how efficient models can extend AI capabilities.
Key points
- A LLM based on Llama 2 ran on a Windows 98 PC with a Pentium II processor.
- The model achieved 39.31 tokens per second performance with only 128 MB of RAM.
- The researchers used a modified llama2.c code called llama98.c compiled with Borland C++ 5.02.
- Utilization of BitNet, a ternary weighted architecture, allows running 7 billion parameter models on older hardware.
- EXO Labs supports open source and aims to enable AI access on outdated devices, with the code available on GitHub.
Why it matters
- The project taps into 2025 IT trends emphasizing low-resource computing, AI democratization, and sustainability.
- Historical significance in the evolution of computing, showcasing how older hardware can still be viable for modern applications.
- Growing interest in optimizing AI models for accessibility and affordability reflects broader trends in tech inclusivity.
This innovative use of old hardware for running advanced AI models illustrates a significant shift towards making AI technology more accessible and sustainable, paving the way for future advancements in computing on legacy systems.