If these were my servers, I would want to see bandwidth usage too, especially if the cloud provider is charging me for it. The PRTG approach looks like a really good option.
I've heard good things about Sensu as well.
But no matter what you use, you need to be able to know what normal operation (performance, capacity, utilization) is like across the servers so you will truly know if the behavior you see is an outlier or expected behavior (i.e. a SQL VM spikes in CPU and memory usage because there are a ton of queries running for order inserts at the end of the day, etc.).