NVIDIA DGX B300
Type: Platform Tags: NVIDIA, DGX B300, Blackwell Ultra, DGX, GB300, AI factory, enterprise AI Related: NVIDIA-DGX, NVIDIA-DGX-B200, NVIDIA-DGX-BasePOD, NVIDIA-DGX-Enterprise-Support, NVIDIA-GB300-NVL72, NVIDIA-DGX-SuperPOD, NVIDIA-DGX-SuperPOD-B300-Spectrum-4-Ethernet-RA, NVIDIA-DGX-SuperPOD-B300-Quantum-X800-InfiniBand-RA, NVIDIA-DGX-Station, NVIDIA-Blackwell-Architecture, NVIDIA-Mission-Control, NVIDIA-AI-Enterprise, NVIDIA-Spectrum-X, NVIDIA-Spectrum-X-Validated-Solution-Stack, NVIDIA-Quantum-X800-InfiniBand, NVIDIA-BlueField-DPU, NVIDIA-Enterprise-AI-Factory Sources: https://www.nvidia.com/en-us/data-center/dgx-b300/, https://docs.nvidia.com/dgx/dgxb300-user-guide/introduction-to-dgxb300.html, https://docs.nvidia.com/dgx-superpod/reference-architecture/scalable-infrastructure-b300/latest/index.html, https://docs.nvidia.com/dgx-superpod/reference-architecture/scalable-infrastructure-b300-xdr/latest/index.html, https://www.nvidia.com/en-us/data-center/gb300-nvl72/, https://www.nvidia.com/en-us/data-center/dgx-superpod/, https://www.nvidia.com/en-us/data-center/dgx-basepod/, https://www.nvidia.com/en-us/data-center/dgx-support/ Last Updated: 2026-05-09
Summary
NVIDIA DGX B300 is NVIDIA’s DGX system generation built around Blackwell Ultra/GB300 infrastructure. It extends NVIDIA-DGX-B200 and GB200 NVL72 into a newer AI factory platform for large training, inference, and enterprise AI deployments, and it is now the basis for current DGX SuperPOD B300 reference architectures.
Detail
Purpose
DGX B300 gives organizations a supported NVIDIA system path for Blackwell Ultra AI factories. It combines dense GPU compute with NVLink, NVIDIA networking, storage partnerships, system software, and operations tools.
Platform role
- DGX-branded system path for Blackwell Ultra.
- Related to NVIDIA-GB300-NVL72 for rack-scale NVLink deployment.
- Deployable within NVIDIA-DGX-BasePOD and NVIDIA-DGX-SuperPOD-style enterprise AI factory patterns, including NVIDIA-DGX-SuperPOD-B300-Spectrum-4-Ethernet-RA and NVIDIA-DGX-SuperPOD-B300-Quantum-X800-InfiniBand-RA.
- Deployment target for NVIDIA-AI-Enterprise, NIM, NeMo, Dynamo, and enterprise AI applications.
- Operated in large deployments with NVIDIA-Mission-Control, BaseOS, Base Command Manager, DCGM, and fabric/storage tools.
Hardware notes
- DGX B300 systems use eight NVIDIA B300 Blackwell Ultra GPUs.
- The DGX B300 user guide lists 8 x 288 GB GPU memory for 2.3 TB total GPU memory.
- Feature summary includes 72 PFLOPS FP8 training and 144 PFLOPS FP4 inference.
- Networking includes eight OSFP ports connected to eight ConnectX-8 cards for cluster networking at up to 800 Gb/s InfiniBand/Ethernet, plus two dual-port NVIDIA-BlueField-DPU BlueField-3 DPUs for storage and management networks at up to 400 Gb/s.
- The docs list DGX OS 7 based on Ubuntu 24.04 LTS, with additional Ubuntu and Red Hat Enterprise Linux support.
NVIDIA context
Use this page for the DGX product/system identity. Use NVIDIA-GB300-NVL72 for rack-scale NVL72 architecture and NVIDIA-Blackwell-Architecture for GPU architecture context.
Connections
- NVIDIA-DGX - DGX B300 is part of the DGX systems family.
- NVIDIA-DGX-B200 - DGX B200 is the previous Blackwell DGX system generation.
- NVIDIA-DGX-BasePOD - enterprise reference architecture path for DGX deployments.
- NVIDIA-DGX-Enterprise-Support - support and services layer for DGX B300 operations.
- NVIDIA-GB300-NVL72 - rack-scale Blackwell Ultra system adjacent to DGX B300 deployments.
- NVIDIA-DGX-SuperPOD - SuperPOD provides the scale-out AI factory pattern.
- NVIDIA-DGX-SuperPOD-B300-Spectrum-4-Ethernet-RA - Spectrum-4/DC-busbar SuperPOD design for DGX B300.
- NVIDIA-DGX-SuperPOD-B300-Quantum-X800-InfiniBand-RA - Quantum-X800/AC-power SuperPOD design for DGX B300.
- NVIDIA-DGX-Station - local GB300 deskside development counterpart.
- NVIDIA-Blackwell-Architecture - Blackwell Ultra extends the Blackwell architecture family.
- NVIDIA-Mission-Control - operations platform for current DGX/NVL72 environments.
- NVIDIA-AI-Enterprise - production AI software stack for DGX systems.
- NVIDIA-Spectrum-X and NVIDIA-Spectrum-X-Validated-Solution-Stack - Ethernet AI factory fabric and validated stack for B300 deployments.
- NVIDIA-Quantum-X800-InfiniBand - 800 Gb/s InfiniBand option in B300 SuperPOD designs.
- NVIDIA-BlueField-DPU - DPU path for storage and management networking.
- NVIDIA-Enterprise-AI-Factory - DGX B300 is an AI factory infrastructure building block.
Source Excerpts
- NVIDIA’s DGX B300 page positions the system as a Blackwell Ultra DGX platform for AI factory-scale workloads.
- The DGX B300 user guide lists eight B300 GPUs, 2.3 TB total GPU memory, ConnectX-8 cluster networking, BlueField-3 storage/management networking, and DGX OS 7 based on Ubuntu 24.04 LTS.