Super computer

Description

This contract award notice informs interested operators about the outcome of a tender. On 12 May 2018, a contract notice was published as part of the EU project PPI4HPC for this public procurement of innovative solutions (PPI). The PPI procurement purchases innovative solutions that address the following challenge: providing high-performance IT equipment for European supercomputing centres. The procurement involves four public procurers (BSC, JUELICH, CINECA and GENCI) located in four different countries (Spain, Germany, Italy and France) working together in a joint procurement. The purpose of the procurement is for each public procurer to purchase an innovative high-performance supercomputer and/or an innovative high-performance storage system that will be integrated in their computing centre. These systems are expected to be deployed in the time frame 2019-2021. In this lot, BSC will acquire storage and computing infrastructure for high-performance data analytics to be installed during the second half of 2019. The BSC HPDA infrastructure will combine compute nodes with innovative storage technologies like NVRAM in conjunction with storage technologies already in place at BSC, such as normal hard drives and tape infrastructure, configured in a tiered storage solution to store hundreds of petabytes of scientific data. For the storage part, an innovative tiered storage solution scalable to hundreds of petabytes is expected. It will feed and provide data to the different BSC HPC and HPDA resources now and in the years to come during the pre-exascale period. For the compute part, a high-performance data analytics infrastructure will be acquired to complement installed BSC HPC production clusters. They will pre- or post-process simulation data with new big data or analytics paradigms and algorithms. Innovative solutions for massive data storage and tiering are expected for this lot. Additional information under this section and the contract notice in its entirety are available on the PPI4HPC website at the following address: https://ppi4hpc.eu For lot 1, one economic operator was selected, as it was the only that presented to this lot. The dialogue phase started with a first context meeting on 23 November 2018 and end of January 2019 as deadline to provide the first offer. The first session of dialogue took place on 1 March 2019, providing feedback on the first offer received. During this meeting a new deadline was set to the company to provide next offer: April 1. The second session of dialogue took place on 7 May 2019, and the final technical specifications were delivered to the company for providing their final offer(s) with a deadline to deliver its final offer before 7 June 2019. Mid July 2019 the evaluation of the final offer was finished, allowing to start with the awarding process. Awarding process start of September. Contract was finally signed on 18 September 2019. Contract implementation will consist of a deployment phase (which is expected to be completed 3 months after contract award) and an operational validation phase of 6 months — during which the solutions will be evaluated in real-life operational conditions). Evaluation will be done by BSC during deploy phase. The goal of the lot is to procure a high-performance computing resource that is planned to be installed at Jülich Supercomputing Centre’s (JSC) supercomputing facility by the end of 2020. The system will replace the currently operated JURECA cluster and must be able to execute the diverse workloads of this system with significantly higher performance and energy efficiency. In addition, the system shall have architectural features specifically targeted at data-intensive science. In particular, JSC expects the system to integrate dense memory technologies into its memory hierarchy and to expose this memory to applications through novel APIs. The system is expected to be connected to the central storage subsystem in the JSC facility with high bandwidth. In addition to the dense memory integration, JSC expects innovative hardware and software solutions offering high energy efficiency as well as improved facility integration (e.g. improved power monitoring and control) and application performance monitoring capabilities. Additional information under this section and the contract notice in its entirety are available on the PPI4HPC website at the following address: https://ppi4hpc.eu The goal of the lot is to procure a new supercomputing system that will replace the scale-out partition of MARCONI. With the deploy of the new system CINECA aims to reach a total computing power of the order of 50 PFlops featuring node architecture that maximise the performance throughput per energy unit. The procured resource integrated in the MARCONI infrastructure represents a new step towards the architectural convergence between high performance computing, high performance data analytics and A.I. and will be able to run computing and data intensive workloads by exploiting the most recent GPU accelerators, high IOPS disks and state-of the-art low latency interconnection. The procurement process lasted ten months and saw five economic operators qualify for this lot, but only four participated to the context meeting in March 2019 due to the withdrawal of one participant. Two subsequent meetings were set, on the 29 April 2019 and 19 June 2019, in order to review the technical roadmaps of the qualified candidates. To evaluate the TCO impact of the proposed solutions, CINECA released to all economic operators the benchmark instruction document and all the benchmark suite, on the 17 June 2019 through public online repository. The final offers, containing the committed results of the benchmarks and based on the final RFP requirements were submitted by three economic operators due to the withdrawal of one qualified candidate. The evaluation panel has opted to require further details to verify the congruity of the offered solutions with respect to the power constraints and the benchmark results, before submitting the final evaluation ranking. Contract implementation will consist of a deployment phase (which is expected to be completed 4 months after contract award) and an operational validation phase of 1 month — during which the solutions will be evaluated in real-life operational conditions). The evaluation phase will be conducted by CINECA staff specialists and selected external users, together with contractor’s engineers, with the aim to properly tune the system for production. The goal of the lot is to increase, during the second half of 2019, the computing power installed in the TGCC (Très Grand Centre de Calcul du CEA, located near Paris) and make it available to French and European researchers. Targeting several tens of Petaflop/s, the new HPC resource will be seamlessly integrated into the system installed in early 2018 in the same location. The new system will provide a balanced architecture for scientific applications, including very high IO bandwidth connection to the storage system of the TGCC. In order to fulfil such requirements and to achieve a high-energy efficiency together with an excellent performance/TCO ratio towards exascale systems, GENCI expects innovative solutions both in terms of hardware (processors, interconnection network...) and in terms of software (novel programming models....) as well as strong support from the provider, particularly during early operation of the new system including a contribution to application porting and optimisation. Additional information under this section and the contract notice in its entirety are available on the PPI4HPC website at the following address: https://ppi4hpc.eu The duration of this framework agreement is due to the availability of the equipment and facilities, the technical complexity of the solution to be implemented and the depreciation period of the equipment. Contract implementation will include the installation of the new system followed by an acceptance phase consisting of a provisional acceptance phase and a pre-production qualification phase with the goal of checking that the system is performing as expected during an early operation stage. The pre-production qualification phase will last 3 months. An operational validation phase lasting 9 months will follow during which the solutions will be evaluated in real-life operational conditions. The selected operators will retain ownership of the intellectual property rights (IPRs) that they generate in the PPI and will be able to use them to exploit the full market potential of their innovative solutions i.e. beyond this procurement. For reference, the worldwide market size of the HPC server markets was evaluated by Hyperion Research to USD 6.7 billion for the first half of 2019.