NVIDIAGB300 NVL72
Rack-scale system built for the AI inference era
All-liquid-cooled rack-level architecture integrates 72 NVIDIA Blackwell Ultra GPUs and 36 Arm-based Grace CPUs on a single platform, delivering 1440 PFLOPS FP4 compute power, 20TB of GPU memory, and 130 TB/s NVLink bandwidth—purpose-built for test-time scaling inference and AI inference workloads.
For AI Inference Performance andDesigned for efficiency
GB300 NVL72 delivers 1.5x denser FP4 Tensor Core performance and 2x attention performance versus Blackwell GPU. AI factories accelerated by GB300 NVL72, powered by Quantum-X800 InfiniBand or Spectrum-X Ethernet, ConnectX-8 SuperNIC, and Mission Control management, achieve up to 50x overall throughput compared to Hopper-based platforms.
TechnologyBreakthrough
From inference acceleration to high-capacity memory, from high-speed networking to vertical scaling interconnects, GB300 NVL72 delivers comprehensive capabilities for large-scale AI inference.
AI Inference Acceleration
Blackwell Ultra Tensor Core delivers 2x acceleration for attention layers and 1.5x AI floating-point compute (vs. Blackwell GPU), purpose-built for test-time scaling inference and AI inference workloads—balancing quality of service with throughput performance.
High-capacity HBM3E VRAM
GPU VRAM capacity increased by 1.5x, supporting larger batch sizes and high-throughput operations to significantly boost AI inference throughput in ultra-long context scenarios.
ConnectX-8 SuperNIC High-Speed Networking
The IO module features two built-in ConnectX-8 adapters, delivering 800 Gb/s network connectivity per GPU. It supports Quantum-X800 InfiniBand or Spectrum-X Ethernet and delivers exceptional RDMA performance.
Fifth-Generation NVLink and Grace CPU
Fifth-generation NVLink scale-up interconnect unlocks accelerated performance for AI inference models; Grace CPU delivers exceptional performance and memory bandwidth, offering energy efficiency up to 2 times that of mainstream server processors.
NVIDIA GB300 NVL72Technical Specifications
Complete parameter overview. Contact our technical consultant for selection advice.
| Settings | 72 NVIDIA Blackwell Ultra GPUs, 36 NVIDIA Grace CPU |
| NVLink Bandwidth | 130 TB/s |
| Quick Memory | 37 TB |
| GPU VRAM | Bandwidth | 20 TB | Up to 576 TB/s |
| CPU & Memory | Bandwidth | 17 TB LPDDR5X | 14 TB/s |
| CPU Cores | 2,592 Arm Neoverse V2 cores |
| FP4 Tensor Core | 1440 PFLOPS (sparse) | 1080 PFLOPS (non-sparse)* |
| FP8/FP6 Tensor Core | 720 PFLOPS |
| INT8 Tensor Core | 24 POPS |
| FP16/BF16 Tensor Core | 360 PFLOPS |
| TF32 Tensor Core | 180 PFLOPS |
| FP32 | 6 PFLOPS |
| FP64 / FP64 Tensor Core | 100 TFLOPS |
| Thermal Performance and Form | Full liquid-cooled rack-level architecture featuring NVIDIA MGX racks. |
| Network | ConnectX-8 SuperNIC: 800 Gb/s per GPU network connectivity Supports NVIDIA Quantum-X800 InfiniBand or Spectrum-X Ethernet |
| Manage software | NVIDIA Mission Control |
* Unless otherwise noted, Tensor Core specifications are for sparse values; 1080 PFLOPS is for dense values. Specifications sourced from the official NVIDIA product page; actual configurations may vary by unit.
Product BrochureDownload
Below is a preview of the NVIDIA GB300 NVL72 product brochure. Click the button to generate and download the PDF version for internal review and selection archiving.
GB300 NVL72 features a full liquid-cooled rack-level architecture that integrates 72 Blackwell Ultra GPUs and 36 Grace CPUs on a single platform, delivering 1440 PFLOPS FP4 compute (sparse), 20TB of GPU memory, and 130 TB/s of NVLink bandwidth. Dense FP4 performance is up to 1.5× higher than Blackwell GPUs, attention mechanism performance improves by 2×, and overall AI factory throughput reaches up to 50× that of the Hopper platform.
| Settings | 72x Blackwell Ultra GPU + 36x Grace CPU |
| FP4 Tensor Core | 1440 PFLOPS (sparse) | 1080 PFLOPS (non-sparse) |
| FP8/FP6 Tensor Core | 720 PFLOPS |
| GPU VRAM | Bandwidth | 20 TB | Up to 576 TB/s |
| Quick Memory | 37 TB |
| CPU Memory | Bandwidth | 17 TB LPDDR5X | 14 TB/s |
| NVLink Bandwidth | 130 TB/s (5th Gen) |
| Number of CPU cores | 2,592 Arm Neoverse V2 cores |
| Network | ConnectX-8 SuperNIC (up to 800 Gb/s per GPU) |
| Thermal Design | Full Liquid-Cooled Rack-Level Architecture · MGX Rack |
Brochures are generated in real-time from page content. If your browser blocks the download, allow pop-ups and try again.
Ready to deploy GB300 NVL72?
Huaying Technology provides end-to-end professional services for GB300 NVL72 selection consulting, liquid-cooled data center planning, and deployment implementation.
Purchase Intent
Submit your requirements and contact details. A sales consultant will confirm pricing, stock availability, and delivery options with you. Product prices are subject to quotation.