Riker, the OLCF’s data analysis and visualization cluster, features 128 CPU nodes, each with dual 64-core AMD EPYC processors and 2.2 terabytes of memory, along with eight GPU nodes, each with one 64-core AMD EPYC processor, 1.5 terabytes of memory, and dual NVIDIA L40S GPUs with 48 gigabytes of GDDR6 memory.
Although Riker’s nodes may be fewer in number than its predecessor Andes’ 704 CPU nodes (each with two 16-core AMD EPYC processors and 256 gigabytes of memory) and nine GPU nodes (transplanted from its predecessor, Rhea, and composed of dual 14-core Intel Xeon processors, 1 terabyte of memory, and two NVIDIA K80 GPUs with 24 gigabytes of GPU memory each), they are far more powerful. Riker uses fourth- and fifth-generation AMD EPYC processors as compared with Andes’ second-generation units. Moreover, Riker’s GPUs feature an up-to-date design in multiple areas, including tensor cores and improved support for mixed-precision operations (which are commonly used in AI and other modern workloads) as well as advances in visualization performance.
