NVIDIA GB300 NVL72
Type: Platform Tags: NVIDIA, GB300 NVL72, Blackwell Ultra, rack-scale AI, NVLink, DGX, AI factory Related: NVIDIA-DGX-B300, NVIDIA-DGX, NVIDIA-DGX-SuperPOD, NVIDIA-DGX-SuperPOD-B300-Spectrum-4-Ethernet-RA, NVIDIA-DGX-SuperPOD-B300-Quantum-X800-InfiniBand-RA, NVIDIA-Blackwell-Architecture, NVIDIA-GB200-NVL72, NVIDIA-NVL72-AI-Factory, NVIDIA-Mission-Control, NVIDIA-Spectrum-X, NVIDIA-Spectrum-X-Validated-Solution-Stack, NVIDIA-Quantum-X800-InfiniBand, NVLink, NVIDIA-Enterprise-AI-Factory Sources: https://www.nvidia.com/en-us/data-center/gb300-nvl72/, https://www.nvidia.com/en-us/data-center/dgx-b300/, https://www.nvidia.com/en-us/data-center/hgx/, https://docs.nvidia.com/enterprise-reference-architectures/nvl72-ai-factory-with-gb300-nvl72-dual-plane-networking-architecture.pdf, https://docs.nvidia.com/networking/software/spectrumx-solution-stack/index.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 Last Updated: 2026-05-09
Summary
NVIDIA GB300 NVL72 is NVIDIA’s Blackwell Ultra rack-scale AI system configuration with 72 Blackwell Ultra GPUs connected as a single high-bandwidth NVLink domain. It follows the NVIDIA-GB200-NVL72 pattern while increasing memory and performance for large-model training and inference.
Detail
Purpose
GB300 NVL72 targets AI factories that need rack-scale GPU memory, GPU-to-GPU bandwidth, and dense inference throughput. It is the Blackwell Ultra step between the original GB200/B200 generation and the later NVIDIA-Vera-Rubin platform.
Public positioning
- Rack-scale NVL72 system form factor built for large-scale AI.
- Uses GB300 Blackwell Ultra generation GPUs and Grace CPU pairing.
- Fits DGX B300, DGX SuperPOD, HGX B300, and OEM/ODM rack-scale system directions.
- The NVIDIA-NVL72-AI-Factory Enterprise RA treats GB300 NVL72 as the rack-scale building block for 72-GPU scalable units and multi-rack Spectrum-X designs.
- Requires an operations and software stack around NVIDIA-Mission-Control, NVIDIA-AI-Enterprise, NVIDIA-DGX-SuperPOD, and AI factory networking/storage.
- NVIDIA-Spectrum-X-Validated-Solution-Stack tracks the current GB300 Spectrum-X software/firmware compatibility signal.
- SuperPOD-scale Blackwell Ultra choices are represented by NVIDIA-DGX-SuperPOD-B300-Spectrum-4-Ethernet-RA and NVIDIA-DGX-SuperPOD-B300-Quantum-X800-InfiniBand-RA for DGX B300, and by NVIDIA-NVL72-AI-Factory for GB300 NVL72 Enterprise RA.
NVIDIA context
Treat GB300 NVL72 as the canonical Blackwell Ultra rack-scale system page. Use NVIDIA-Blackwell-Architecture for architecture-level details and NVIDIA-DGX-B300 for the DGX-branded system page.
Connections
- NVIDIA-DGX-B300 - DGX B300 is NVIDIA’s DGX system path around Blackwell Ultra.
- NVIDIA-DGX - DGX systems are the broader product family.
- NVIDIA-DGX-SuperPOD - GB300 NVL72 fits SuperPOD-scale AI factory designs.
- NVIDIA-DGX-SuperPOD-B300-Spectrum-4-Ethernet-RA and NVIDIA-DGX-SuperPOD-B300-Quantum-X800-InfiniBand-RA - adjacent B300 SuperPOD architecture choices.
- NVIDIA-Blackwell-Architecture - GB300 is Blackwell Ultra, extending the Blackwell family.
- NVIDIA-GB200-NVL72 - predecessor rack-scale Blackwell NVL72 system.
- NVIDIA-NVL72-AI-Factory - Enterprise RA for GB300 NVL72 rack-scale AI factory deployments.
- NVIDIA-Mission-Control - operations layer for modern DGX/NVL72 environments.
- NVIDIA-Spectrum-X and NVIDIA-Spectrum-X-Validated-Solution-Stack - Ethernet AI factory fabric and validated GB300 stack path.
- NVIDIA-Quantum-X800-InfiniBand - InfiniBand option for large B300/Blackwell Ultra SuperPOD fabrics.
- NVLink - rack-scale NVLink is the defining interconnect.
- NVIDIA-Enterprise-AI-Factory - enterprise AI factory guidance includes this class of validated infrastructure.
Source Excerpts
- NVIDIA’s GB300 NVL72 and DGX B300 pages position GB300 as a Blackwell Ultra rack-scale AI system for large AI training and inference.