Don't have Telegram yet? Try it now!
https://hgpu.org/?p=17418
Optimized Broadcast for Deep Learning Workloads on Dense-GPU InfiniBand Clusters: MPI or NCCL?