AI High-Performance Server: Delivery, Installation, and Commissioning

Delivery of an AI high-performance server of the NVIDIA DGX B200 performance class or a technically equivalent system (HGX B200 or similar): including installation, commissioning, and ready-for-operation handover at the point of use. The server is intended for computationally intensive AI tasks (post-training of large language models), further development, fine-tuning, and expansion of the sovereign language models (Soofi-S 30B and 120B) developed in the SOOFI project, including industrial adaptation in the DAISEC environment. Required are a total GPU memory of over 1 TB and high computing power; integration into an existing compute cluster is necessary. The service includes the fully functional overall system, including GPUs, CPUs, network components, and memory/RAM. A manufacturer's warranty of at least 24 months for all hardware components is included, with extension options to 36 or 60 months, callable until the end of the current warranty. The service includes the integration of the complete system, a preliminary system test, and assembly and commissioning. Licenses for NVIDIA AI Enterprise are not part of the tender. An optional configuration with 4 TB DDR5 ECC system RAM instead of 2 TB is possible.

Tender documents

Additional details