“ReThink GPU: Building Energy-Efficient AI Systems"

Inspiration

SAM ALTMAN mentioned about the Heating/melting of the GPUs during Ghibli art creation in a joking way but I think it's going to be a major problem in the upcoming days. So, due to the rapid growth of AI technologies, there is high energy consumption, overheating of GPUs and E-waste. Listening about how the data centers and AI workloads strain the environment inspired me to explore Sustainable AI computing solutions. I wanted to create AI based Thermal monitoring system that monitors and predicts GPUs thermal load and dynamically redistributes processing tasks, contributing t greener technology

What it does

The AI based monitoring system basically monitors and predicts GPUs thermal load and dynamically redistributes processing tasks that uses intelligent cooling and load balancing to prevent overheating

How we built it

firstly, we started analyzing the major environmental impacts of GPU usage/ overuse of GPU that leads to E-waste, high energy consumption and heat generation. so, we started researching on sustainable hardware like NPU, Photonic processors using the simulation and modeling tools we thought that we could design an AI based workload management system that predicts GPU thermal load and redistributes tasks dynamically to reduce overheating and power loss. We also explored optimized chip architectures and the use of eco-friendly materials for sustainable hardware design.

Challenges we ran into

First, it was difficult to understand how GPUs actually cause heating and power loss during AI processing. We had to research about how GPU load, temperature, and energy use are connected. Another challenge was finding simple ways to make GPUs more energy-efficient without affecting their speed. Learning about new technologies like NPUs and photonic chips was exciting but tricky, since most information is very advanced. Despite these challenges, we worked together and tried to find solution

Accomplishments that we're proud of

As a 2nd year ECE student and the team who is concerned about environmental issues We’re proud that we were able to clearly understand how GPUs work and how their heavy use affects the environment. This project helped us connect what we learned in electronics with real-world issues like energy saving and sustainability. We came up with an idea to make AI systems more efficient by managing GPU workload and reducing heat generation. We also learned about new and exciting technologies like NPUs, neuromorphic, and photonic processors. Most importantly, we turned a big global problem into a simple, practical concept that students like us can explain and build on.

What we learned

Through this project, we learned how GPUs play a major role in AI and how their high-power use affects the environment. We also understood the importance of making technology not just faster, but also energy-efficient and eco-friendly. We also learned how to research and analyze technical topics like GPU heating, workload management, and alternative processors such as NPUs and photonic chips. This helped us see how electronics and computer science come together in real-world applications. most importantly how to make this work

What's next for “ReThink GPU: Building Energy-Efficient AI Systems"

This is the idea we have we have to focus on making it into a prototype. we would try to create a small model using Arduino or raspberry pi to show how the GPU heat and power can be monitored thorugh sensors and can be controlled using Arduino We also would love to study more and gain knowledge about new energy-efficient chips like NPUs and photonic processors. In the future, we’d like to develop simple software that can predict GPU load and adjust power automatically to save energy. Our goal is to make AI systems that perform well while being kind to the environment. With more research, guidance, and teamwork, ReThink GPU can grow into a real-world solution for sustainable computing.

Built With

• arduino
• easyeda
• python
• pytouch
• tensorflow