System Architecture

NRP is a heterogenous, nationally distributed, open system that features CPUs, FP32- and FP64-optimized GPUs, and FPGAs, arranged into two types of subsystems ("high-performance" at SDSC and two "FP32-optimized" at UNL, MGHPCC), specialized for a wide range of data science, simulations, and machine learning or artificial intelligence, allowing data access through a federated national-scale content delivery network (CDN). The NRP HPC system features a novel extremely low-latency fabric from GigaIO that allows dynamic composition of hardware including FPGAs, GPUs, and NVMe storage. Each of the three sites (SDSC, UNL, and MGHPCC) includes ~1 PB of useable disk space. The three storage systems function as data origins of a content delivery network (CDN) that will provide data access anywhere in the country within an RTT of ~10ms via use of network caches. NRP's data infrastructure supports a national "Bring Your Own Resource" (BYOR) program for campuses to add additional compute, data, and storage resources to NRP. The system also can be scaled out via "Bring Your Own Device" (BYOD) programs.

NRP HPC Subsystem (located at SDSC)

NRP FP32 Subsystem (1 each at UNL and MGHPCC)