Util - Compute Cost & Carbon Optimization Engine

Built as part of a broader research and product initiative exploring cost and carbon optimization in AI and data center infrastructure.

Demo

https://utilplatformv01.streamlit.app/

Website for download (in progress):

https://utilcompute.com/

Overview

Util is a decision-support tool that optimizes when and where computational workloads should run to minimize electricity cost, carbon emissions, or a weighted combination of both.

The system models compute as a constrained optimization problem under real-world conditions by using the variables:

• time-varying electricity prices
• carbon intensity of the grid at any given time
• Workload time requirements
• deadline constraints
• Historical data

Util produces an optimal execution schedule along with quantified cost and emissions outcomes.

Why this matters

As compute demand grows (AI training, data centers, research modeling, GPU workloads), energy consumption is becoming a primary cost and sustainability issue.

Key Features

• Optimization Engine
  • Minimizes cost, carbon, or a weighted objective
  • Built using constrained optimization
• Workload Scheduling
  • Allocates compute across time intervals
  • Respects users deadlines and compute requirements
• Forecast Integration
  • Uses time-series electricity price and carbon intensity data along with machine learning to determine an estimate for future costs and emissions beyond public provided information
• Balanced Optimization Mode
  • Allows trade-offs between cost and emissions
• Detailed Outputs
  • Optimal schedule (run vs pause)
  • Cost and emissions breakdown
  • Savings vs baseline
  • CSV output
• Interactive Interface
• Workload Shifting (in progress)
  • Determines which data centers have access to the cleanest or cheapest energy and shift workloads accordingly

System Architecture

User Input ↓ Location Mapping ↓ Energy Data Fetcher ↓ Forecast Engine ↓ Optimization Engine ↓ Scheduler ↓ Dashboard Output

Tech Stack

Python pandas / numpy — data processing scipy.optimize — optimization engine Streamlit — frontend interface Altair — visualization

How It Works

Util solves: Minimize: Cost = Σ(power × price × time) or Carbon = Σ(power × carbon_intensity × time) Subject to: required compute hours deadline constraint machine capacity

Installation

1. Clone the repository git clone https://github.com/yourusername/util.git cd util

2. Install dependencies pip install -r requirements.txt

3. Create a .env file in the project root The app expects .env at the root of the running workspace, next to app.py. If you run Util in a mounted or containerized workspace, that workspace must also contain its own .env file. Do not commit .env to GitHub.

Template:

WATTTIME_USERNAME=your_watttime_username
WATTTIME_PASSWORD=your_watttime_password
AWS_ACCESS_KEY_ID=your_aws_access_key_id
AWS_SECRET_ACCESS_KEY=your_aws_secret_access_key
AWS_REGION=your_aws_region
S3_BUCKET_NAME=your_private_s3_bucket_name

For Streamlit Community Cloud, add AWS credentials in the app Secrets settings instead of committing .env. Expected secret names:

• AWS_ACCESS_KEY_ID
• AWS_SECRET_ACCESS_KEY
• AWS_REGION
• S3_BUCKET_NAME

Example Streamlit secrets format:

AWS_ACCESS_KEY_ID = "your_aws_access_key_id"
AWS_SECRET_ACCESS_KEY = "your_aws_secret_access_key"
AWS_REGION = "your_aws_region"
S3_BUCKET_NAME = "your_private_s3_bucket_name"

4. Run the app streamlit run app.py

Use Cases

• AI model training optimization
• GPU workload scheduling
• data center cost reduction
• carbon-aware compute planning
• research and academic compute workloads

About

Util is being developed as an early-stage product exploring the intersection of: infrastructure economics energy systems cost optimization AI compute

Author

Finn Case Economics @ UC Santa Barbara Focus: infrastructure, energy systems, and cost optimization in technology environments