1. Researchers from UC Santa Cruz have discovered a method to run large language models (LLMs) at a mere 13 watts without compromising performance. 2. The key to this efficiency is the elimination of matrix multiplication in LLM processing, which, when optimized, significantly boosts performance-per-watt. 3. The broader applicability of this approach to AI in general is yet to be determined.
Related Articles
- A Lighter, Smarter Magnetoreceptive Electronic Skin2 months ago
- Designing to Support Energy-Efficient Edge AI in Process Applications2 months ago
- Storage Research Framework: The Development Path of DRAM Technology2 months ago
- Hannover Messe: Smart Materials Offer Sustainable Cooling and Heating Solutions for Homes, Cars and Industry3 months ago
- Hannover Messe: New Climate Technology for Home, Car, and Industry – How Wires and Sheets Cool and Heat Sustainably3 months ago
- Exclusive: Huawei Noah's Ark Lab Director姚骏 Succeeded by Youngest Leader Wang Yunhe3 months ago
- Hannover Messe: Pumps and valves made from ultrathin elastomeric films are lightweight and energy efficient3 months ago
- Hannover Messe: Smart, energy-efficient robot grippers cut production costs3 months ago
- Thin Chips and Robust Substrates - Key Technologies for Cost-Efficient Silicon Carbide Power Electronics3 months ago
- Outstanding Final Theses Awarded by Baelz Foundation: HBC Graduate Johannes Spangler Receives Prize3 months ago