Benchmarking AI Systems: From Model Metrics to Real-World Performance

Abstract

AI benchmarking is evolving rapidly as enterprises scale from experimentation to deployment. This interactive session explores measuring real world performance across inference and training workloads. We will discuss metrics that matter, throughput vs. latency tradeoffs, memory bandwidth, and open software ecosystems. Gain practical insights into evaluating AI infrastructure for performance, scalability, efficiency, and TCO in modern enterprise and developer environments.