Design a monitoring strategy to track GPU utilization, memory bandwidth, and network bottlenecks across a cluster of 128 GPUs.