NVIDIA GB300 NVL72 Liquid-Cooled Rack System
Model / Platform:GB300 NVL72
GB300 NVL72 is delivered as a rack, integrating 72 NVIDIA Blackwell Ultra accelerators and featuring DLC-2 direct liquid cooling. Pre-integrated and validated at the rack level before shipment, it helps customers shorten deployment cycles for large-scale AI infrastructure.
Key Highlights
Technical Specifications
| Delivery Format | Rack-scale |
|---|---|
| Accelerator | 72 × NVIDIA Blackwell Ultra |
| Cooling | DLC-2 Direct Liquid Cooling |
| Network | Pre-integrated high-speed interconnects within racks and external networks |
| Power | Rack-Level Power and Busbar Solutions |
| Deploy | Pre-integrated at factory; quick cabinet assembly on-site |
Specifications are based on public family-level information from Supermicro. Specific parameters may vary by configuration; final selection and pricing are subject to confirmation and inquiry results.
Typical Applications
Download
The product datasheet includes key features, specifications, typical application scenarios, and selection recommendations. Download it for your records or internal evaluation.
Typical Use Cases and Selection Recommendations
CombineGB300 NVL72platform positioning, common deployment scenarios, and key selection criteria to streamline requirement alignment during evaluation and quoting.
Typical Use Cases
Selection Recommendation
Deployment Environment
Liquid cooling integration and data center retrofitting require completing site surveys and power distribution planning in advance.
Configuration Orientation
Plan compute, power, and networking at the rack level to avoid single-server selection.
Complementary & Extensions
Requires integration with a cooling capacity distribution unit and secondary-side piping, along with planning for routine liquid cooling system operations and maintenance.
The above is a general selection reference. Specific configurations, compatibility, and pricing are subject to confirmation upon inquiry.
Compare specs for models in the same category
SelectGPU ServerCustomize comparisons for the models below. Parameters highlighted in blue differ from the current model. Export a comparison PDF for selection and record-keeping.
Ultra-Small Form Factor (USFF) Specification Comparison Sheet
Compare Models:GB300 NVL72, SYS-82 1GE-TNHR, AS-412 5GS-TNRT, ARS-11 1GL· Category:GPU Server· Generated:2026/10/26
| Specification Item | Current DeviceNVIDIA GB300 NVL72 Liquid-Cooled Rack SystemGB300 NVL72 | 8-Card NVIDIA H200 / H100 Air-Cooled GPU ServerSYS-821GE-TNHR | 4U AMD EPYC GPU ServerAS-4125GS-TNRT | NVIDIA GH200 Grace Hopper MGX SystemARS-111GL |
|---|---|---|---|---|
| Delivery Format | Rack-scale | — | — | — |
| Accelerator Card | 72 × NVIDIA Blackwell Ultra | 8 × NVIDIA H200 / H100 SXM | 8 × Dual-width GPU or 12 × Single-width GPU | — |
| Cooling | DLC-2 Direct Liquid Cooling | Air-cooled | Air-cooled | — |
| Network | Pre-integrated high-speed interconnects within racks and external networks | High-speed Ethernet / InfiniBand with DPU support | — | High-Speed Ethernet / InfiniBand |
| Power | Rack-Level Power and Busbar Solutions | — | — | — |
| Deploy | Pre-integrated at factory; quick cabinet assembly on-site | — | — | — |
| Product Form | — | 8U Rack-Mount | 4U Rack | 1U Rack-Mounted (MGX Architecture) |
| Processor | — | Dual Intel Xeon Scalable Processors | Dual AMD EPYC 9004 Series | — |
| memory | — | DDR5, multi-channel high capacity | DDR5, multi-channel | High-bandwidth LPDDR5X and HBM |
| Expansion slots | — | — | PCIe 5.0 x16 | — |
| Calculation Module | — | — | — | NVIDIA GH200 Grace Hopper Superchip |
| Interconnect | — | — | — | NVLink-C2C High-Bandwidth Consistent Interconnect |
| Typical Scenarios | — | — | — | Scientific Computing, AI Training & Inference |
Table compares only official family-level public specifications; missing items are marked with "—". Final configuration and pricing are subject to confirmation upon inquiry.
Related products in the same category
ViewGPU ServerAll Products
SuperServer GPU
8-Card NVIDIA H200 / H100 Air-Cooled GPU Server
8U Air-Cooled Multi-GPU Training Platform for Large Model Training and Inference Deployment in Traditional Data Centers.
A+ Server
4U AMD EPYC GPU Server
4U Dual-Socket AMD EPYC GPU Platform supports up to 8 dual-width or 12 single-width GPUs, balancing compute power with expansion flexibility.
MGX Modular System
NVIDIA GH200 Grace Hopper MGX System
Grace Hopper Superchip System based on NVIDIA MGX architecture, designed for high-bandwidth scientific computing and AI workloads.
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.