NVIDIA GB200™ NVL72
The pinnacle of the NVIDIA Blackwell platform, a liquid-cooled architecture designed for trillion-parameter Large Language Models.
NVIDIA GB200 NVL72 connects 36 Grace CPUs and 72 Blackwell GPUs to build a massive rack-scale supercomputer. Through liquid cooling and 5th Gen NVLink, it functions like a single GPU, providing unprecedented real-time inference and training capabilities for multi-trillion parameter models.
Architecture Highlights
Extreme Compute
World-leading NVIDIA Blackwell architecture, accelerating trillion-parameter model training.
Liquid Cooling
Full rack integrated liquid cooling for higher energy efficiency and thermal performance.
Ultra-Fast Interconnect
Featuring fifth-generation NVLink for massive data exchange between GPUs.
Application Scenarios
Optimized for variety of advanced computational workloads
Trillion-Parameter Training
Providing superior parallel compute for LLMs, significantly shortening training cycles.
High-Fidelity Simulation
Accelerating intensive simulations like fluid dynamics, molecular modeling, and weather forecasting.
Large-Scale Inference
Supporting real-time, high-throughput AI inference with extreme energy efficiency.
Core Capabilities
Technical Specifications
| GPU Configuration | 72x Blackwell GPU |
| Total Memory | 13.5 TB HBM3e |
| Total Bandwidth | 130 TB/s |