Impact & Results
The Carbon Aware AI Workload Manager and SLURM pseudo-scheduler projects have significantly advanced the sustainability of AI and machine learning systems. The Carbon Aware AI Workload Manager helps AI researchers estimate and reduce the carbon impact of their workloads, while the SLURM pseudo-scheduler enables developers to optimize training based on renewable energy availability and track greenhouse gas emissions. These tools support Intel’s sustainability goals and foster more environmentally friendly machine learning practices.
Problem Statement
AI and machine learning workloads contribute significantly to greenhouse gas emissions, yet developers lack the specialized tools needed to estimate, manage, and reduce their environmental impact. The absence of solutions for optimizing training based on energy efficiency and renewable energy availability leads to inefficient practices, making it challenging for developers to align their workflows with sustainability goals.
Process & Approach
I designed and led the development of the Carbon Aware AI Workload Manager to estimate the carbon impact of AI workloads and offer options for reduction. Working with Intel Labs, I integrated their research with input from our internal developers and researchers to create a SLURM pseudo-scheduler. This tool helps machine learning developers optimize training times and locations based on renewable energy availability while also tracking emissions.
After deployment I created several videos for demos presented at various internal forums and conferences.
Tools & Methods
Miro
Excel
Grafana
Final Cut Pro
Jira
Miro used for collaboration between teams
Excel for structure and data
Built in Grafana to provide visualizations
User testing conducted
Demo creation for various conferences & communication purposes
All work was tracked in Jira