numaperf studies how topology-aware scheduling affects latency-critical AI workloads. It implements NUMA-first memory allocation and thread placement strategies in Rust, measuring the performance impact of hardware-conscious resource management.