The carbon footprint of computers, the Internet, and its infrastructure is growing and it is comparable to the amount of the carbon emissions produced by the airline industry globally as of today. Reaching net-zero – neither a surplus nor a deficit of carbon emissions – imposes fundamental challenges and rethinking on how we design and construct computer systems for the future. As our world is becoming increasingly digitalized, there is an exponential increase in data to store, process, and explore, often involving the training and use of sophisticated AI models. This is driving the need for ever more compute and storage capacity, leading to a greater amount of energy demand and carbon emissions.

This Dagstuhl Seminar will bring together experts from computer science and computer engineering that share a common vision for power and energy-efficient computing. Five principal topic areas will be discussed:

Reaching Net-Zero in the Age of AI and Machine Learning

• How do we reduce the barriers to designing CO2-efficient AI systems, considering trade-offs between model precision and carbon emissions?
• How can we build tools that assess the operational carbon footprint of software?
• How do we efficiently utilize new technologies (e.g., non-volatile memory) to reduce the CO2 footprint of computing systems?

Carbon Transparency from Source to Sink

• Can we (and how) extend existing energy modeling to carbon modeling and execution (workloads and architectures), including computation, storage, and communication?
• What is the carbon emission information needed by toolchains (e.g., carbon “debuggers” or monitoring), and how to communicate it most efficiently and effectively?
• Can other environmental aspects such as water consumption, chemical waste, and noise emissions be derived from carbon emission metrics?

Carbon-aware Computing, Storage, and Communication

• How can we model the time-varying and spatially distributed carbon intensity of running applications at each location, including the cost of moving the data, computations, and results?
• Can we model the geospatial matching of energy consumption to renewable generation? How fine-grained must our geospatial awareness be?

Disruptive Paradigms

• How can we adapt to carbon-modulated computing and new hardware capabilities?
• What are the Wild And Crazy Ideas (WACI) to explore future carbon-neutrality?

Carbon-aware Computing Hackathons

• What are the practical techniques to reduce emissions? What are the suitable metrics?
• What are the pre-requisites, scope, and limits of each technique and does it transfer to other platforms/environments?

Creative Commons BY 4.0

Kerstin I. Eder, Timo Hönig, Maja Hanne Kirkeby, Daniel Mosse, and Max Plauth