Through the center for optimized modern parallel adaptive system software (COMPASS), improve modern systems with complex CPU, accelerator, memory, and network architectures

The Department of Energy National Nuclear Security Administration awarded the University of New Mexico a cooperative agreement valued at $1,555,556 on September 22, 2025, under the Predictive Science Academic Alliance Program (CFDA 81.124) to develop and deploy the Center for Optimized Modern Parallel Adaptive System Software (COMPASS), which uses AI-enhanced performance models to improve design and runtime performance of parallel applications, runtimes, and system software across modern systems with complex CPU, accelerator, memory, and network architectures.

The award funds a five-year research and development effort spanning September 22, 2025, through September 21, 2030, performed in Albuquerque, New Mexico. During the initial two years, center personnel will deploy infrastructure, assess limitations and risks in existing state-of-the-art approaches, and evaluate performance of integrating AI-enhanced methods into a coherent system. Years three and four focus on advancing the state of the art through iteratively refined implementations of architecture components, with results informing more complex, integrated, and distributed models and optimizations.

The University of New Mexico, through its Office of Sponsored Projects Division, serves as the prime recipient. PSAAP emphasizes verification, validation, and uncertainty quantification methodologies for large-scale simulations, with focus on leveraging high-performance computing to advance predictive science and build long-term recruiting pools for the NNSA laboratory complex.

The COMPASS approach to addressing these challenges is based on the systematic use of AI-enhanced performance models to improve the design and runtime performance of parallel applications, runtimes, and system software. Over the five-year span of COMPASS, center personnel will iteratively research, develop, deploy, and assess this architecture and its individual components. Research in the first two project years focuses on deploying infrastructure, assessing limitations and risks in existing state-of-the-art approaches, and assessing the performance of integrating these methods into a coherent system. Years three and four will then focus on significantly advancing the state of the art across the research thrusts described in the following sections through iteratively refined implementations of architecture components, with results from this research informing more complex, integrated, and distributed models and optimizations.