Germany - IT Services: Consulting, Software Development, Internet and Assistance

Description

The object of the procurement is an HPC cluster with a partitionable high-speed network.The cluster consists of nodes with server CPUs, and nodes with several GPUs. The quantity ratios will be determined during the negotiation process.The computers will be operated either with air cooling or with direct water cooling (possibly with partial (<30%) air cooling).In case of air cooling, the computers will be installed by the contractor in existing Rittal racks (120cm deep, 60cm wide) with Rittal LCP 60KW sidecooler, the used space will be included in the total cost with a virtual price. The racks have a built-in power distribution of 2x3x32A, in case of higher demand a further distribution has to be retrofitted.In case of direct water cooling the contractor supplies own racks with water distribution and water/water heat exchanger. For each 90cm space there is a DN50 cold water supply (12°C) and hot water return (>18°C) with a pressure difference of approx. 1500 hPa. The system must also be convertible to operation with a supply of 35°C and a return >40°C after a possible later conversion to a new computer room. The racks must be equipped with a power distribution system that measures the power consumption at least per phase; the operator must be able to read out these measured values electronically at least 4 times per minute. The cluster has 2 networks: all nodes are connected to the IPMI network for hardware management and monitoring with their baseboard management controllers (BMC) via a network connection reserved only for the BMC. The switches of the IPMI network use the "private vlan" feature to prevent the nodes from communicating with each other over the IPMI network.In addition, all cluster nodes and have connections to a high-speed network. The bandwidth of each connection is communicated in the framework with the tender documents. This high-speed network is partitionable with software commands into at least 4 individual partitions that cannot communicate with each other. This control will be controlled from only one of the partitions, the control software will write an audit log of the configuration.The current clusters of the ZDV are operated with "CentOs", "xcat" and "slurm", therefore a compatible solution is preferred.The maximum energy consumption of the offered solution is less than 650KW in all operating conditions.Selection criterion will be a formula for evaluating the total cost of performance, which relates the performance of the cluster in given criteria to the total cost (creation and operation) over a number of years. The energy efficiency of the overall system will be included as a factor in the evaluation formula. The exact formula and performance criteria will only be disclosed to the selected bidders with the bills of quantities.The cluster will be delivered with "next working day on-site service", a more detailed specification will be made during the negotiation process.
Between 4 and 8 participants will be selected from the applicants, who will receive the invitation to tender together with the detailed specifications. For further details, please refer to the specifications.

Opportunity closing date

04 October 2021

Value of contract

£5m-50m