# New Technique Could Cut AI Energy Use by 95%! ⚡

**URL:** <https://onehack.st/t/new-technique-could-cut-ai-energy-use-by-95/293163>\
**Category:** News & Articles\
**Tags:** technology\
**Created:** [October 9, 2024, 4:32pm UTC](https://onehack.st/t/new-technique-could-cut-ai-energy-use-by-95/293163 "2024-10-09T16:32:11Z")\
**Posts on this page:** 1\
**Page:** 1

<div class="post-metadata">

**Author:** ![TheJoker](https://onehack.st/user_avatar/onehack.st/thejoker/32/22660_2.png) [@TheJoker](https://onehack.st/u/TheJoker)\
**Post date:** [October 9, 2024, 4:32pm UTC](https://onehack.st/t/new-technique-could-cut-ai-energy-use-by-95/293163/1 "2024-10-09T16:32:11Z")

</div>

**Summary:**

1. **Breakthrough Technique**  
Researchers at **BitEnergy AI** have introduced **Linear-Complexity Multiplication (L-Mul)**, a method that can reduce AI model energy consumption by up to **95%**.

2. **How It Works**  
L-Mul replaces complex floating-point multiplications with simpler integer additions, making calculations faster and more energy-efficient without losing accuracy.

3. **Significant Energy Savings**  
By using L-Mul, the energy cost of floating-point tensor multiplications can drop by **95%** , while energy for dot products can be reduced by **80%**.

4. **Minimal Performance Trade-off**  
Tests across various AI tasks show only a **0.07%** average performance drop, indicating that the energy savings come with negligible sacrifices in accuracy.

5. **Hardware Requirements**  
To fully exploit L-Mul’s benefits, specialized hardware is necessary. Plans for developing hardware that supports L-Mul calculations are underway, aiming to unlock its full potential.

**Read more at:** [Decrypt](https://decrypt.co/285154/new-technique-slashes-ai-energy) | [arXiv Research Paper](https://arxiv.org/pdf/2410.00907)