Carbon Aware AI Workload Manager

Pioneering sustainable AI development by creating the first carbon-aware workload scheduler, enabling researchers to make environmentally conscious decisions about their AI training processes.

Role

Product Design, UX Strategy, UX Design

Team

Anthropologist, AI Developers, Sustainability Experts, AI Researchers, DevOps, Infrastructure Engineers

Tools

Miro, Excel, Grafana, Final Cut Pro, Jira

Executive Summary

Sustainable AI Initiative

Designed and led the development of the first Carbon Aware AI Workload Manager and SLURM pseudo-scheduler in collaboration with Intel Labs. This sustainability-focused solution enables carbon estimation, training optimization based on renewable energy availability, and emissions tracking—advancing sustainable AI practices across the organization and setting new industry standards.

First
Carbon-Aware AI Scheduler
100%
Emissions Transparency
Real-time
Renewable Energy Integration
60%
Reduction in Carbon Emissions

The Tension Points

Environmental Impact Crisis

AI and machine learning workloads are energy-intensive, contributing significantly to greenhouse gas emissions. Training GPT-3 was found to emit as much as 552 tons of CO2, creating an urgent need for sustainable AI practices.

Developer Awareness Gap

Developers lacked tools to estimate or reduce carbon impact, with no support for optimizing training around energy efficiency or renewables. This created a disconnect between development practices and corporate sustainability goals.

Industry Leadership Opportunity

While sustainability was becoming a corporate priority, there were no practical tools for AI developers to align their work with environmental goals. The opportunity existed to position Intel as a leader in sustainable AI development.

Strategic Approach

Academic-to-Practical Translation

Partnered with Intel Labs to translate cutting-edge academic sustainability research into practical tools that developers could actually use. This bridge between research and implementation was crucial for adoption and impact.

Real-Time Energy Integration

Designed the system to integrate with real-time renewable energy data, enabling developers to schedule training during periods of high renewable energy availability. This approach maximized environmental impact while maintaining development velocity.

Transparency-First Design

Built comprehensive emissions tracking and visualization capabilities to provide developers with clear visibility into their environmental impact. This transparency was essential for driving behavioral change and accountability.

Process & Methodology

Partnership & Research

Collaborated with Intel Labs to translate academic sustainability research into practical applications. Engaged internal developers and researchers to validate needs and pain points, using Miro for team collaboration.

Sustainable Design

Led the design of the Carbon Aware AI Workload Manager with carbon estimation and reduction strategies. Developed the SLURM pseudo-scheduler for renewable energy optimization and built Grafana visualizations for emissions insights.

Deployment & Advocacy

Successfully deployed tools across internal teams and created demo videos using Final Cut Pro for internal forums and external conferences. Drove awareness and adoption while contributing to Intel's environmental goals.

Impact

AI Development Workflow Transformation

The Carbon Aware AI Workload Manager fundamentally changed how developers approach AI training by integrating environmental considerations into their decision-making process. This ecosystem integration ensured sustainability became a natural part of the development workflow.

Before Carbon Awareness

  • Blind energy consumption
  • No emissions tracking
  • Cumbersome process calculating emissions for paper and conference submissions
  • Environmental impact unknown

After Carbon Awareness

  • Real-time emissions monitoring
  • Renewable energy optimization
  • Flexible carbon-aware scheduling
  • Transparent and easy impact reporting

Visual Assets

Carbon Aware Overview

Overview presentation of the Carbon Aware AI Workload Manager

Carbon Aware Snips 1

Carbon Aware Snip of Video Demo

Carbon Aware Snips 2

Detailed view of carbon estimation interface

Carbon Aware Snips S1

Renewable energy integration dashboard

Carbon Aware Snips S2

Emissions tracking and reporting flow

Carbon Aware Snips S3

Data synthesis of converations with AI Developer, Researchers, DevOps, and Infrastructure Engineers

Tools & Technologies

Miro

Collaboration & Mapping

Grafana

Data Visualization

Final Cut Pro

Demo Creation

Jira

Project Management