NVIDIA DGX B300

Specifications

Overview

The NVIDIA DGX B300 is the first DGX system built on NVIDIA Blackwell Ultra architecture, delivering 1.5× the dense FP4 performance and 2× the attention performance of its predecessor DGX B200. With 8× Blackwell Ultra GPUs, 2.1TB of HBM3e memory, and 14.4 TB/s of GPU interconnect bandwidth, the DGX B300 is purpose-built for the AI reasoning era — enabling enterprises to run trillion-parameter models, agentic AI workflows, and high-throughput inference serving at hyperscale efficiency. First DGX system deployable in NVIDIA MGX racks, setting a new standard for data center AI infrastructure.

Key Benefits

AI Reasoning Engine: Purpose-built for LLM inference and training workloads that demand massive attention compute — 2× attention throughput vs DGX B200 enables real-time reasoning for agentic AI. MGX-Compatible: First DGX deployable in NVIDIA MGX modular racks alongside traditional enterprise racks, simplifying procurement and deployment. Sustainable Performance: Multiple power configuration options deliver highest performance-per-watt in its class. Full-Stack Software: NVIDIA AI Enterprise, NIM microservices, and CUDA-X libraries included — deploy AI factories in days, not months.