Nvidia Hgx B200
NVIDIA HGX™ B200
High-performance server platform purpose-built for the Blackwell architecture.
The NVIDIA HGX B200 combines eight Blackwell GPUs interconnected via fifth-generation NVLink to provide unprecedented compute density. It is the most powerful single-node solution for generative AI and LLM inference today.
Architecture Highlights
15x Inference Perf
Overwhelming advantage in LLM inference compared to previous generations.
1.1TB Memory
Massive HBM3e capacity to support the largest models.
Application Scenarios
Optimized for variety of advanced computational workloads
Generative AI Training
Rapid iteration of GPT-scale and similar large pre-trained models.
Real-time LLM Inference
Serving millions of users with low-latency content generation.
Core Capabilities
2.5 PFLOPS FP8 Performance
144GB HBM3e Memory
1.8TB/s NVLink Bandwidth
Superior Energy Efficiency
Technical Specifications
| GPU Quantity | 8x Blackwell B200 |
| Compute (FP8) | 36 PFLOPS |
| Total Memory | 1152 GB |