GPU Use Case

Graphics Processing Units (GPU) are recently the biggest source of computing power. They were developed as a specialized tool for processing of images mainly in gaming. But currently they are applied in a growing number of CPU heavy applications: not only related to picture processing but also to artificial intelligence and others.

In Earth Observation (EO) domain GPUs are being applied, among others to radar interferometry. In the following case the user is processing Sentinel-1 SLC radar data in a medium size GPU VM computing setup and stores the results into the object storage. GPUs are installed in the so called single server VMs. These are the cloud Virtual Machines that fill the full physical server and no other VMs can reside there. These physical machines are additionally equipped with a GPU card – and this card can be reached by a user with so called “passthrough” mechanism. This mechanism is a way to minimalise the overhead of a cloud computing hypervisor.

In this example a user uses: Sentinel-1 SLC data for radar interferometry, 4 dedicated server GPU VMs: dsgp.large and 100 TB of object storage – all in a 1 year contract:

Figure 1 - GPU Use Case Table