Accelerated Computing Hardware
NVIDIA GB200™ NVL72 Rack Scale Solution
The pinnacle of the NVIDIA Blackwell platform, liquid-cooled architecture designed for trillion-parameter Large Language Models.
NVIDIA GB200 NVL72 connects 36 Grace CPUs and 72 Blackwell GPUs into a single massive unified GPU. Engineered with fifth-generation NVLink, it enables trillion-parameter LLM training and real-time inference at 30x lower energy and cost.
Key Architecture Highlights
- Perfect synergy of 36 Grace CPUs and 72 Blackwell GPUs in a single rack
- Up to 30x inference performance boost compared to prior generation H100
- Full rack integration with advanced liquid cooling and power distribution
- Official sourcing, logistics, and certified on-site acceptance testing
Official Hardware & Procurement Assurance
Supplied directly by XCLOUD NETWORK TECH CO., LTD. with full manufacturer warranty, official serial numbers, and Cambodia/international delivery logistics.
Technical Specifications
| Architecture | NVIDIA Blackwell + Grace CPU |
| Total GPUs / CPUs | 72 Blackwell GPUs + 36 Grace CPUs |
| AI Performance | Up to 1,440 PFLOPS (FP4) / 720 PFLOPS (FP8) |
| Memory | 13.5 TB Fast HBM3e Memory |
| NVLink Bandwidth | 130 TB/s Bidirectional Scale-up |
| Cooling Design | Liquid-Cooled 100% Direct-to-Chip |
| Power Requirement | 100 kW - 120 kW Rack Standard |
Need Custom Configuration for NVIDIA GB200™ NVL72 Rack Scale Solution?
Our engineers can assist with thermal planning, power estimation, and network switch matching.