Don't have Telegram yet? Try it now!
https://hgpu.org/?p=7913
Optimizing All-to-All and Allgather Communications on GPGPU Clusters