ESCADE - Energy-Efficient Large-Scale Artificial Intelligence for Sustainable Data Centers
Monday, May 13, 2024 3:00 PM to Wednesday, May 15, 2024 4:00 PM · 2 days 1 hr. (Europe/Berlin)
Foyer D-G - 2nd floor
Project Poster
AI Applications powered by HPC TechnologiesEmerging Computing TechnologiesHeterogeneous System ArchitecturesML Systems and ToolsSustainability and Energy Efficiency
Information
Poster is on display.
The power consumption of data centers has doubled in the last ten years and will account for 13% of global energy consumption by 2030. AI is one of the biggest drivers of data center power consumption: training a natural language processing model such as the GPT-3 model consumes 936,000 kWh and generates approximately 284 t of CO2. ESCADE's goal is to significantly reduce the energy requirements of data centers by using world-leading hardware and software technologies to improve the environmental footprint of AI applications. The focus will be on the use of neuromorphic chip technologies, as these promise efficiency gains of up to 50% in training and up to 80% in the inference of AI models. The SpiNNcloud data center at TU Dresden based on neuromorphic SpiNNaker2 chips will serve as prototype for evaluation of two use-cases: Visual computing for steel industry and efficient training of language models for digital industry. An AI sustainability framework will be developed to monitor the sustainability of AI systems going far beyond mere power measurement. Concepts for integrating neuromorphic chips into classic GPU-based data centers will help planning more sustainable AI data centers. This makes a concrete contribution to decoupling economic growth and prosperity from resource consumption.
Contributors:
The power consumption of data centers has doubled in the last ten years and will account for 13% of global energy consumption by 2030. AI is one of the biggest drivers of data center power consumption: training a natural language processing model such as the GPT-3 model consumes 936,000 kWh and generates approximately 284 t of CO2. ESCADE's goal is to significantly reduce the energy requirements of data centers by using world-leading hardware and software technologies to improve the environmental footprint of AI applications. The focus will be on the use of neuromorphic chip technologies, as these promise efficiency gains of up to 50% in training and up to 80% in the inference of AI models. The SpiNNcloud data center at TU Dresden based on neuromorphic SpiNNaker2 chips will serve as prototype for evaluation of two use-cases: Visual computing for steel industry and efficient training of language models for digital industry. An AI sustainability framework will be developed to monitor the sustainability of AI systems going far beyond mere power measurement. Concepts for integrating neuromorphic chips into classic GPU-based data centers will help planning more sustainable AI data centers. This makes a concrete contribution to decoupling economic growth and prosperity from resource consumption.
Contributors:
Format
On-site
