The purpose of the Invitation to Tender is to select one (1) economic operator (hereafter “Contractor”) for the component acquisition, delivery, assembly, hardware and software installation and maintenance of an AI-ready EuroHPC machine (“Supercomputer”) to establish an associated “AI Factory” and that will be owned by the EuroHPC JU. The EuroHPC JU will conclude one contract with the successful tenderer in order to acquire, deliver, install and maintain the High-End Supercomputer hosted by the respective Hosting Entity. Therefore, the prospective Contractors should have proven experience of acquiring, delivering, installing and maintaining Supercomputers in similar environments as the ones required in this procurement procedure. The solutions to be procured are expected to outperform existing solutions regarding several of the following aspects: • Performance/TCO; • Programmability and usability; • Versatility; • System stability; • Power and energy efficiency; • Computing density.

GPU & CPU partition and full flash storage

A high-performance GPU-accelerated computing partition to be delivered as part of the Discoverer ++ GPU & CPU Partition. The solution is designed to support AI workloads, large-scale simulations, and data-intensive research across a wide spectrum of scientific and industrial domains. LOT 1 should contain two different GPU compute infrastructure with fast interconnect and one AI optimized storage with direct GPU access feature.

Wafer-Scale AI Accelerator System, Specialized AI servers, FPGA servers, Cloud servers.

The cluster must offer at least two AI accelerator chips based on -wafer scale, integrating hundreds of thousands of cores into a single silicon wafer, designed for large-scale deep learning workloads. The Specialized AI Servers must assume the delivery and installation of homogeneous cluster with height density at least 32 AI accelerators or more with very high speed (based on the same hardware and software configuration) compute nodes (servers). This tender specification defines the requirements for procuring FPGA AI-accelerated systems for the BRAIN++ AI Factory at Sofia Tech Park. Each server unit must be configured to host and efficiently operate 4x FPGA accelerator cards, designed for high-performance computing workloads with emphasis on memory-intensive applications.

Additional UPS system

A high availability N+1 UPS solution with 2N(N+1) for the most critical loads.