Carbon ∅

A blockchain network for task-sharing across idle computers to cut waste and carbon emissions.

Comment

Inspiration

As fullstack developers who have worked on a number of AI projects in the past, we have always been mindful of how heavy AI training tasks are on our computers, as well as its impact on the environment. There are a few key statistics that inspired us to build an emissions-reducing solution:

  • 77% of the time, computers are in idle or low-intensity modes, using only ~1% of their GPU
  • 20 million metric tons of CO2 could be saved annually in the US if computers reduced their energy usage by 30%
  • $19+ billion is the yearly cost of idle electronics in the US, with over 50% of that coming from consumer devices specifically

In light of the US government’s recent shift away from ESG programs amidst the ~30% compound annual growth rate of the AI industry, we wanted to create something that could fill that gap. That’s why we created Carbon ∅.

What it does & how we built it

Carbon ∅ is a compute resource-sharing network which distributes loads to the most underutilized computers, increasing energy efficiency while democratizing access to AI development resources. It promotes collaboration and enables everyone to develop smarter AI’s with a smaller footprint.

We developed the desktop application with Electron and networked through socket.io-client. The backend HTTP server was developed with Express, websockets were implemented with Socket.IO, and we stored the queue of jobs in a custom BlockChain implementation.

Accomplishments that we're proud of & what we learned

Over the past 24-hours, we’ve learned a lot. To list just a few things…

  • Working a lot with low-level process execution
  • Making a hybrid, Python and JavaScript project for the first time
  • Building the largest project we’ve ever completed in a hackathon in just 24 hours

What's next for Carbon ∅

Scalability Testing: stress-testing the queueing system across larger pools of idle machines
Incentive Mechanism: implement a token or credit system to reward frequent contributors
Partnerships: pilot with universities, research labs, or enterprises that need AI training power

Share this project:

Updates