Enabling Macroheterogeneity Through a System of Systems Approach

Learn about implementing macroheterogeneity in computing systems through a comprehensive "system of systems" approach in this 15-minute conference talk from the Open Compute Project. Explore how moving beyond traditional homogeneous system designs can unlock significant value for AI co-design applications, where diverse computational phases can be optimally mapped to specialized compute partitions including modeling-simulation, inference, and data analysis segments. Discover how AI surrogates can strategically replace application components to enhance inference query capabilities while maintaining system performance. Examine the critical framework requirements for successful system-of-systems implementation, including unified approaches to fabrics, data services, resource scheduling, runtimes, and system management that ensure sustained high utilization across heterogeneous components. Understand the architectural flexibility needed to support both loose and tight integration between partitions based on varying communication requirements. Gain insights into real-world use cases that demonstrate the practical applications of system-of-systems architectures, learn about the key architectural challenges that must be addressed, and explore the current tools and methodologies available to overcome these obstacles. Analyze performance potential through relevant application examples that showcase the tangible benefits of this approach to modern computing infrastructure design.