Compute servers are the primary computational resource in an HPC cluster. These servers are responsible for running the preferred workloads. Optimal compute configuration depends on the applications and customer specific goals. There are several configurations available including heterogenous options. Heterogenous configurations are popular as many environments have mixed workloads. For example, in a traditional HPC environment, many applications are CPU only. And an organization that focuses on traditional HPC plus some deep learning would benefit from tailoring the configuration towards dense compute with a few GPU-based compute servers.

Table 2 lists the available compute platforms. The options vary in density and compute type (GPU support). These include AMD-based and Intel-based options, the selection of which depends on customer needs and workload.

Table 2.      Compute server configuration options