The NVIDIA DGX B300 System is a next-generation enterprise AI supercomputer designed for AI reasoning, Generative AI, LLM training/fine-tuning, and high-throughput inference. Built on NVIDIA Blackwell Ultra, it delivers massive GPU memory, ultra-fast GPU-to-GPU connectivity, and modern scale-out networking—making it a powerful foundation for building an AI Factory in the data center.
Key Highlights
- 8× NVIDIA Blackwell Ultra SXM GPUs for accelerated GenAI and reasoning workloads
- 2.1 TB total GPU memory (massive capacity for long-context models and larger batch sizes)
- NVLink Switch System (2×) + 14.4 TB/s aggregate NVLink bandwidth for fast multi-GPU scaling inside the system
- AI performance (as listed by NVIDIA): FP4 Tensor Core up to 144 PFLOPS; FP8 Tensor Core 72 PFLOPS
- Networking: 8× OSFP ports for 8× NVIDIA ConnectX-8 VPI, up to 800 Gb/s InfiniBand/Ethernet for scale-out clusters
- Enterprise host platform: Intel® Xeon® 6776P processors
- Data-center ready power/thermal design: ~14.5 kW power consumption; 10RU form factor
