HPC Iridis 6

Tender description

The University is now looking to procure an updated system (IRIDIS 6) to replace IRIDIS4 under the SSSNA Framework; Lot 4: Solutions - High Performance Computing (HPC) and Data Intensive Computing (DIC) Only. It is anticipated that the new system will have a high-level specification as follows (illustrated in Figure 1 below): Schematic illustrating draft specification of required system. • Management infrastructure with highly recoverable management nodes. • At least 3 login nodes, each with a minimum of 2TB of RAM and 10 TB of RAID-enabled storage. • Compute nodes arranged in such a way to support highly coupled parallel jobs of up to 32 node. • A total number of standard compute nodes which maximizes the budget available with at least 4GB of RAM per core and 250 GB of local storage. • At least 2 'high-memory' nodes with at least 1.5 TB of RAM and 250 GB of local storage. • A management network with a performance at least matching that of 1Gbps Ethernet. • A high-performance, low-latency network with a bandwidth of at least 100 Gb/s. The system will be hosted in the University's data centre. The data centre is specifically designed for water-cooled racks and has limited air-cooling capacity. Therefore, the supplied system should aim to exhaust all waste heat to water. The maximum allocated power load for the system is 300 kW. Note that no high-performance storage is required; this will requirement will be managed under a separate procurement activity. Provision must be made for the HPC system to interface with such storage.