Server system for artificial-intelligence inference workloads with eight GPUs

Summary

The tender covers the purchase of one rack-mounted server system for artificial-intelligence inference workloads, with eight GPUs each providing more than 280 GB of HBM memory, together with five years of next-business-day on-site support. The system must include a 128-core AMD Epyc 9755 processor or equivalent, 2 TB of DDR5 memory, eight NVMe SSDs of approximately 15 TB each, two system SSDs of at least 512 GB, 25 Gbit/s and 100 Gbit/s network connections, and remote console access. The system must be delivered by 28 February 2027 at the latest. Hardware is paid against a 100% bank guarantee; the invoice and guarantee must be available by 11 December 2026. Support is paid annually in advance with a 30-day net payment term. The total contract price carries 75% of the evaluation weight and system performance 25%. Performance scoring covers CPU capacity and GPU throughput for mixture-of-experts models. The bidder must provide a complete technical and commercial offer, a company profile, turnover data for the last three fiscal years, exclusion-ground declarations, a sanctions declaration, liability insurance evidence, subcontractor information where applicable, and a signed readiness declaration for the required advance-payment guarantee.