NCCL collective latency tables: all_reduce tops median latency at 24.5µs, peaks at 839 Gbps
StasBekman · x · 2026-10-15
Stas Bekman shared benchmark tables of collective communication latency and bus bandwidth.
- Latency (small payloads): broadcast and reduce are fastest at 12.3µs; allreduce is slowest at 24.5µs with just 16B.
- Bandwidth (16GiB payloads): allreduce peaks highest at 839.56 Gbps; alltoall is lowest at 639.74 Gbps, with half-peak payload sizes listed per collective.
Useful reference for anyone optimizing distributed training or inference communication.
More from Infra
- NVIDIA and Sweden's Wallenberg launch Sferical One, a shared AI supercomputer — nvidia · 2026-10-16
- Next infra frontier: self-evolving systems that adapt and self-verify — _shreya_s · 2026-10-16
- Cerebras pushes disaggregated inference: heterogeneous prefill pools feeding wafer-scale decode — ai · 2026-10-16
- d-Matrix GA's Corsair rackscale, demos first public Attention-FFN disaggregation at 845 tok/s — ai · 2026-10-16
- Actual's inference engine hits 10k tok/s on NVIDIA DGX Station, GB300 case study — natesiggard · 2026-10-16
- TSMC Q3 revenue hits $46.7B, up 41% YoY; Q4 guidance of $52-54B tops estimates — Beth_Kindig · 2026-10-16