Intel® True Scale Fabric Overview
Maximizing scalability for tomorrow’s high performance computing (HPC) platforms
Today’s high performance computing (HPC) environments are more demanding than ever. To meet the needs of complex simulations and ever-increasing massive workloads, HPC clusters need to support faster, high core count processors, coprocessors, and scale to larger node counts that run more demanding applications.
To meet these challenges, an HPC network environment must be designed from the ground up to meet or exceed requirements, not added onto preexisting general purpose software and hardware. Intel® True Scale Architecture supports the most demanding HPC applications. The Intel® True Scale Fabric Host Channel Adapters (HCA) along with its host stack, Performance Scaled Messaging (PSM), were specifically developed to provide an optimized, connectionless transport mechanism for each server. Together with the Intel® True Scale Fabric Switch 12000 Series, a very high-performing fabric for HPC clusters and their Message Passing Interface (MPI) applications is created. Every member of the Intel True Scale Fabric Switch 12000 Series provides predictable, scalable performance, even under heavy traffic and extreme message rate load. All Intel True Scale Fabric products can be managed via the Intel True Scale Fabric Suite, providing an overall view into the health, status, and performance of a truly high performance computer cluster.