
The U.S. National Science Foundation has awarded nearly $600,000 to support a groundbreaking project at UC Merced aimed at accelerating the future of artificial intelligence and scientific research.
Professor Xiaoyi Lu from the electrical engineering and computer science group in the School of Engineering and his team have created the Open Source Data Offloading and Transfer Architecture (OpenDOTA) project. The work focuses on improving how computers equipped with specialized hardware called Data Processing Units (DPUs) move data efficiently. By making data transfer faster, OpenDOTA allows the main computer system to concentrate on important tasks, which can speed up scientific research and the development of artificial intelligence.
The OpenDOTA project will develop technology to help computers move and process massive amounts of data more efficiently, a crucial step for faster discoveries in science and smarter AI systems.
Artificial intelligence is, as one NPR story put it, “an electricity hog.” According to the Allen Institute for AI, a single query to ChatGPT uses roughly the same amount of electricity as lighting one lightbulb for 20 minutes.
AI consumes far more electricity than traditional internet use. As ChatGPT and programs like it grow in popularity, that’s a big increase in the energy needed to power them.
With this investment, UC Merced researchers will develop innovative tools that enhance the efficiency of computers in handling data-intensive AI and scientific tasks, enabling scientists and engineers to concentrate on solving complex problems rather than waiting for data to transfer between systems. The OpenDOTA project is designed to benefit not only academic researchers but also industries that rely on high-performance computing, opening the door to new advancements in fields ranging from agriculture to climate science.
Currently, many computer systems struggle to keep up when transferring large amounts of data, especially in environments where multiple computers work together. OpenDOTA aims to solve these slowdowns by making data movement faster and more efficient. As Professor Lu explains, improving the efficiency of DPU-powered systems will help advance scientific simulations, drive progress in AI, and strengthen the overall computing infrastructure for improved performance and energy consumption. The project encourages collaboration and will be available for anyone to use, supporting both academic and industry needs.
The research is organized around three main goals: intelligent data transfer between computers, efficient group communication among multiple machines and scalable learning-based automated optimization.
“By combining these approaches, OpenDOTA offers a complete solution to today’s data movement challenges, helping to power the next generation of high-performance and sustainable AI and scientific applications,” Lu said.
The project’s website will provide updates and resources for anyone interested. The $596,738 grant, awarded through the National Science Foundation’s Office of Advanced Cyberinfrastructure Core Research, will support this project for three years.