During East-West fabric validation on a 64-GPU cluster, an engineer runs all_reduce_perf andobserves an algorithm bandwidth of 350 GB/s and bus bandwidth of 656 GB/s. What does thisindicate about the fabric performance?
During East-West fabric validation on a 64-GPU cluster, an engineer runs all_reduce_perf andobserves an algorithm bandwidth of 350 GB/s and bus bandwidth of 656 GB/s. What does thisindicate about the fabric performance?