Increasingly, cloud computing is becoming more and more popular to run large-scale algorithms. This provides us with the opportunity to combine look outside and look within strategies. This can be done by provisioning one or more GPUs in multiple virtual machines, as shown in the following screenshot:

Making the best use of hybrid architecture is a non-trivial task. This is approached by first dividing the data into multiple partitions. Compute-intensive tasks that require less data are parallelized within each node at the GPUs.