You can choose how much compute power you want to concentrate in HPC clusters.
But most of this compute power has been wasted because of the reality of the 1 application per server model.
