Deep Learning GPU Cluster

Summary

The tender concerns the procurement of a Deep Learning GPU Cluster based on NVIDIA B300 GPUs for Fraunhofer IFF. The bidder will supply and install on-site, including commissioning. The system comprises 8 NVIDIA Blackwell B300 GPUs (PCIe 5.0), a total GPU memory of 2304 GB, 72 PFLOPS training and 144 PFLOPS inference capability, NVLink interconnect, 128 CPU cores, 2 TB system memory, 30 TB internal storage, and networking up to 800 Gb/s (InfiniBand/Ethernet). The on-site installation, Ubuntu LTS operating system with necessary drivers, and license-free software are included. A 3-year hardware maintenance and on-site support package is provided. The installation will take place at Fraunhofer IFF in Magdeburg. Delivery is planned by 30.04.2027. Payment terms are net 30 days after successful delivery and acceptance. The tender is an open procedure with a single award criterion based on price.