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.
NUMA-first memory allocation and thread placement for latency-critical AI workloads in Rust.
numaperf is one option in a category that includes hwloc, libnuma, Linux sched_setaffinity , and custom NUMA schedulers. Our Compare page has the full side-by-side.