Measuring Sustainable Software: Which Metrics Really Matter

Energy consumption and carbon emissions are only the beginning. Explore the metrics that make software sustainability measurable and comparable.

Sustainability in software is primarily determined by two key indicators: energy consumption and CO₂ emissions. They show how resource-intensive a software system is and what impact its operation has on the environment.

For a comprehensive assessment of software sustainability, however, additional aspects should also be considered. Hardware resource utilization, maintainability, infrastructure efficiency, and AI-related metrics all contribute to the long-term sustainability of a software system. Our article on sustainable software development explores how these considerations can be integrated into the software development process.

This article introduces the most important categories of sustainability metrics and explains which indicators organizations and engineering teams can use to evaluate sustainable software.

Why Do We Need Sustainability Metrics?

Sustainability can only be improved if it can be measured. While traditional software metrics focus on functionality, quality, and performance, sustainability metrics additionally evaluate the environmental, technical, and organizational impact of software.

These include:

• Energy Consumption
• Runtime and Performance
• Hardware and Resource Utilization
• Environmental and Sustainability Metrics
• AI and Machine Learning Metrics
• Maintainability
• Data Center and Infrastructure Metrics

Each of these perspectives highlights a different aspect of sustainable software. Together, they provide a more complete understanding of how efficient, resource-conscious, and future-proof a software system truly is.

But how can these aspects be measured in practice? The following sections introduce the most important categories of sustainability metrics and the key indicators behind them.

Energy Consumption

Energy consumption is the most important sustainability metric for software because it forms the basis for nearly all other environmental indicators. It describes how much electrical energy a software system requires during operation. This includes not only total consumption, but also the distribution across individual components such as processors, memory, and graphics hardware.

To understand energy efficiency, the following questions should be addressed:

• How much energy does a software system consume?
• Which components account for the largest share?
• How efficiently does the software perform useful work?

Typical metrics include:

• Total Energy Consumption – total energy used by a system or application.
• CPU Energy Consumption – energy consumed by the processor.
• GPU Energy Consumption – energy consumed by graphics hardware, particularly relevant for AI workloads.
• Memory Energy Consumption – energy consumed by system memory.
• Power Consumption – instantaneous power usage.
• Energy Efficiency – the relationship between useful work performed and the energy required.
• Power Usage Effectiveness (PUE) – a metric that evaluates data center energy efficiency by comparing total facility energy consumption with the energy used by IT equipment.

These metrics help identify energy-intensive components and uncover optimization opportunities.

Runtime and Performance

Runtime and performance metrics evaluate how quickly and efficiently software executes its tasks. They are not only relevant for user experience but are often directly related to energy consumption.

Key metrics include:

• Runtime / Execution Time – total time required to complete a task.
• Response Time – time between a request and the first system response.
• Latency – delay in processing or transmitting data.
• Throughput – number of transactions, requests, or work units processed per unit of time.

These metrics help identify bottlenecks and assess the efficiency and performance of software systems.

Hardware and Resource Utilization

These metrics evaluate how efficiently software uses available hardware and system resources. They provide detailed insights into software behavior at the operating system and hardware level and help identify optimization potential.

Key metrics include:

• CPU Utilization, GPU Utilization, and Memory Utilization – usage levels of processors, graphics hardware, and memory.
• Cache Misses and Cache References – indicators of memory access efficiency.
• Context Switch Count and System Call Count – metrics related to interaction with the operating system.
• CPU Frequency – processor clock frequency during execution.
• Network Usage, Data Transfer, and Bytes Read/Written – measures of network and storage resource consumption.

These metrics help organizations better understand technical efficiency and reduce both resource consumption and energy demand.

Environmental and Sustainability Metrics

Environmental and sustainability metrics help quantify the ecological impact of software. They support systematic assessment and comparison of software systems in terms of climate impact, resource consumption, and environmental performance.

Key metrics include:

• Operational Carbon Emissions – CO₂ emissions generated during software operation.
• Embodied Carbon Footprint – CO₂ emissions associated with manufacturing, transportation, and disposal of hardware.
• Water Consumption and Water Footprint – water usage directly or indirectly caused by IT systems.
• Green AI Quotient, Green IT Index, and EcoScore – indicators used to assess the sustainability performance of software, AI systems, and IT infrastructures.

These metrics provide transparency into the environmental impact of software systems and enable a more informed sustainability assessment across their lifecycle. Measuring software sustainability throughout its lifecycle is also a focus of the Green IT Impact (GITI) research project.

Software Carbon Intensity (SCI)

SCI is an ISO-accredited standard that measures software carbon emissions as a rate, rather than as totals and offsets. Small portions of the application, called functional units, serve as the basis for calculating this rate at the component level, which is then scaled up to represent the entire software application. Using various functional units that model different aspects of the software provides a more thorough view of its carbon emissions.

AI and Machine Learning Metrics

As artificial intelligence becomes increasingly widespread, specialized metrics are needed to assess model quality, efficiency, and resource requirements. They help relate the performance of AI systems to their technical and environmental costs.

Key metrics include:

• Accuracy, Precision, Recall, and F1 Score – indicators of prediction quality and model performance.
• Model Size, Parameter Count, and Trainable Parameter Count – measures of model complexity.
• FLOPS and MAC Operations – indicators of computational effort.
• Tokens per Second and Tokens per Request – measures of operational efficiency.
• Energy Scaling Factor and Model-to-Memory Ratio – metrics relating resource consumption to model characteristics.

These metrics support a holistic assessment of AI systems and help organizations develop and operate models more efficiently.

Maintainability

Sustainable software should not only run efficiently but also remain maintainable and adaptable over time. Metrics in this category help assess software quality and long-term maintenance effort.

Key metrics include:

• Maintainability – overall maintainability of a software system.
• Cyclomatic Complexity – complexity of source code and execution paths.
• Lines of Code (LOC) – size of the codebase.
• Code Smells and Energy Smells – indicators of quality or efficiency issues in the code.
• Modularity – degree to which software is organized into independent and reusable components.

These metrics help identify technical debt early and improve software sustainability over the long term.

Data Center and Infrastructure Metrics

The sustainability of software is strongly influenced by the efficiency of the underlying infrastructure. In cloud and data center environments, idle resources, overprovisioning, and inefficient utilization can significantly increase energy consumption.

Key metrics include:

• Annual Component Consumption (ACC) – yearly energy consumption of infrastructure components.
• Annual Idle Consumption (AIC) – energy consumed while systems remain idle.
• Relative Idle Consumption (RIC) – share of idle energy consumption relative to total consumption.
• Component Consumption per Unit of Work (CCUW) – energy required per unit of useful work.
• Provisioning (PRO) – ratio of allocated resources to actual usage.
• Operational Overhead (OPO) – additional resource consumption required to operate infrastructure.

These metrics help improve the efficiency of data centers and cloud environments while reducing resource and energy waste.

Which Software Sustainability Metrics Should Companies Prioritize?

The most important metrics for a company are those that correspond to its application. For example, a database-heavy application benefits more from data transfer metrics, while a graphics-heavy application benefits more from performance metrics. We recommend determining the most important metrics based on standard use cases of the application. Then, choose metrics that reflect these use cases. Follow this decision bidirectionally — backward to the design decisions that influence the use case and forward to its consequences.

In other words, use one main metric per use case that corresponds to performance, runtime behavior, or resource efficiency. Then, select fitting design and sustainability impact metrics. For example: Energy Smells -> CPU utilization -> CPU energy consumption -> Operational Carbon Emissions.

This process gives companies a better understanding of how resource- and energy-efficient their software applications are. Then, they can optimize performance and efficiency to reduce their environmental output.

Conclusion

Sustainable software cannot be reduced to a single metric. While energy consumption and the resulting CO₂ emissions remain the most important foundations for sustainability assessments, additional metrics provide a more comprehensive understanding.

Metrics related to runtime and performance, hardware and resource utilization, maintainability, AI systems, and infrastructure help identify optimization opportunities and better understand the long-term impact of software.

By continuously measuring and analyzing these indicators, organizations can make informed decisions, use resources more efficiently, and build more sustainable software systems. After all, only what is measured can be improved. For additional perspectives on sustainable software development and Quality Engineering, explore our German Testing Day 2026 insights.

This article was created with the help of AI.

Sources and further information: ISO/IEC 25010, Green Software Foundation SCI Specification, MLPerf, SPECpower and GHG Protocol.