GPU instances are the instances provided by AWS that works best for applications with massive parallelism e.g. workloads using thousands of threads. Graphics processing is the example that has huge computational requirements, there each of the tasks is relatively small, all the set of operations performed forms a pipeline, and the throughput obtained of this pipeline is most important than the latency of all the individual operations. Being able to build applications that exploit this degree of parallelism, one will need GPU device and specific knowledge by understanding how to program will work against various graphics APIs or the GPU compute programming models.