NVIDIA Blackwell GPU Architecture Delivers 4x AI Training Performance

The Next Generation of AI Compute

NVIDIA has officially launched its Blackwell GPU architecture, marking a significant leap in AI training and inference capabilities.

Key Specifications

• 4x improvement in AI training performance over Hopper
• 30x improvement in inference for large language models
• 25% reduction in power consumption per computation
• Support for models with trillions of parameters

Pricing and Availability

The B200 GPU is expected to ship in volume starting next quarter. Cloud providers including AWS, Azure, and Google Cloud have already committed to large-scale deployments.

Impact on AI Development

This hardware advancement is expected to accelerate the development of even larger and more capable AI models, potentially enabling new capabilities that were previously computationally infeasible.