NVIDIA DGX B200 System

The NVIDIA DGX B200 is the factory-built 8-GPU Blackwell system from NVIDIA itself: a 10U chassis that arrives with the HGX B200 baseboard, two Intel Xeon Platinum 8570 processors, 2 TB of system memory expandable to 4 TB, and the networking already installed. Eight single-port ConnectX-7 adapters give each GPU a dedicated 400 Gb/s InfiniBand or Ethernet path for cluster fabrics, while two BlueField-3 DPUs serve storage and in-band management. The system delivers 72 PFLOPS of FP8 training and 144 PFLOPS of FP4 inference from 1.4 TB of HBM3e.

DGX B200 suits organisations that want a validated, supported appliance rather than an OEM integration project: research institutes, financial firms, media companies and enterprises standing up their first on-premises training cluster. Units are combined into DGX SuperPOD or DGX BasePOD reference designs, and NVIDIA Base Command, DGX OS and NVIDIA AI Enterprise are bundled with the hardware, which shortens the time from delivery to first training run. Facilities planning matters: each system can draw around 14.3 kW and needs matching power distribution and cooling capacity.

For customers in the Gulf and neighbouring markets, InTechnologies FZCO sources DGX B200 systems through the NVIDIA authorised channel. The commercial contract and invoice come from a licensed partner of ours, while InTechnologies manages procurement, staging and consolidation in Dubai and delivers DDP or DDU to the data centre. As a Blackwell-based product, DGX B200 is subject to US export regulations, and the NVIDIA end-user verification process is completed before shipment.