RapidsMPF shuffles data at 1.8 TiB/s across eight GPUs

Accelerated Out of Core Shuffling

RapidsMPF is a reusable out-of-core shuffler that turns shuffling OOM headaches into a spill you can budget for. On a single DGX B200 with eight Blackwell GPUs, it hits roughly 1.8 TiB/s of global throughput. The post explains why shuffles are memory-intensive rather than compute-intensive, walks through distributed hash joins, and benchmarks the C++ bench_shuffle tool across transports and memory limits.