Unrivalled
Performance in..
Generative AI and LLM Development
Accelerate LLM development with up to 144 PFLOPS FP4 Tensor Core performance.
AI Training at Hyperscale Efficiency
Achieve great model throughput with 72 PFLOPS FP8/FP6 performance, ultra-fast 14.4 TB/s NVLink bandwidth and high-density compute.
High-Speed AI Inference for Production Deployment
Deploy real-time inference at scale using the Blackwell Ultra architecture, delivering up to 144 PFLOPS sparse FP4 for advanced reasoning tasks.
Key Features of the
NVIDIA HGX B300
Blackwell Ultra Architecture
The NVIDIA B300 GPU is built on the Blackwell Ultra architecture, so you can handle real-time AI reasoning, massive parallelism and dense compute throughput across FP4, FP6 and FP8.
Massive 2.1 TB GPU Memory Space
8 Blackwell Ultra GPUs deliver a combined 2.3 TB of ultra-high-bandwidth HBM3e, providing 64 TB/s aggregate bandwidth to keep your largest models fully in memory.
NVLink Switch System for Extreme Scaling
The dual NVIDIA NVLink Switch System provides 14.4 TB/s of aggregate GPU interconnect bandwidth for near-linear scaling across all eight GPUs.
