NVIDIA DGX B200
Boost AI with NVIDIA DGX B200
Introducing the groundbreaking NVIDIA DGX B200 GPU, the world's first system powered by the revolutionary NVIDIA Blackwell architecture. This cutting-edge solution delivers unparalleled performance to tackle the most complex AI tasks like generative AI, LLM and NLP.
New Era of Generative AI with NVIDIA DGX B200
Redefine AI Performance
Packed with 208 billion transistors, Blackwell-architecture GPUs are manufactured using TSMC's cutting-edge 4NP process. It features two interconnected GPU dies, forming a unified GPU with a 10 TB/second chip-to-chip link. With the second-generation Transformer Engine and 4-bit floating point AI inference capabilities, Blackwell supports double the compute and model sizes, propelling AI performance to unprecedented heights. The fifth-generation NVLink delivers a mind-blowing 1.8TB/s bidirectional throughput per GPU
Reliability Meets Security
Blackwell's dedicated RAS Engine ensures unparalleled reliability, availability and serviceability, while its AI-based preventative maintenance capabilities maximise system uptime and reduce operating costs. Advanced confidential computing features safeguard AI models and customer data without sacrificing performance. Integrated decompression engine accelerates database queries, delivering unmatched performance in data analytics and data science
Benefits of NVIDIA DGX B200
Higher Inference Performance
Up to 15x faster inference for massive models like GPT-MoE-1.8T compared to the previous Hopper generation. With cutting-edge Blackwell Tensor Core technology, combined with TensorRT-LLM and Nemo Framework innovations, get unmatched acceleration for LLM and Mixture-of-Experts models
Transformer Engine
Experience 3x faster training GPT-MoE-1.8T with the second-generation Transformer Engine, featuring groundbreaking 8-bit floating point (FP8) and new precisions. Fifth-generation NVLink interconnect (1.8TB/s GPU-to-GPU bandwidth), high-speed InfiniBand networking, and NVIDIA Magnum IO software
A New Class of AI Superchip
Built with Blackwell Architecture with 208 billion transistor and a custom-built 4NP TSMC process, it features two-reticle limit GPU dies connected by a blazing 10 TB/second chip-to-chip link, forming a unified GPU of unparalleled power
RAS Engine for Reliability
Equipped with a dedicated engine for reliability, availability, and serviceability (RAS). The latest Blackwell architecture incorporates AI-based preventative maintenance capabilities at the chip level for diagnostics and forecasting of potential reliability issues. Maximise system uptime for massive-scale AI deployments
Secure AI for Confidentiality
Advanced confidential computing capabilities protect your AI models and customer data without compromising performance. New native interface encryption protocols, critical for privacy-sensitive industries like healthcare and financial services
Decompression Engine
Accelerate database queries and experience the highest performance in data analytics and data science with a dedicated decompression engine. Supporting the latest formats, this engine allows you to process vast amounts of data efficiently
FP4 Precision
FP4 precision feature in the groundbreaking Blackwell architecture. This innovative technology delivers a quantum leap in AI performance, enabling you to push the boundaries of what's possible
Petaflop-Scale AI Performance
AI performance with up to 144 petaflops at your disposal. This staggering computational prowess allows you to tackle even the most complex and data-intensive AI challenges with ease
Massive Memory Capacity
1.4TB of GPU memory and 64TB/s of memory bandwidth. With this unparalleled combination of memory capacity and bandwidth, you can handle the largest AI models and datasets for seamless scalability
Supercharge Gen AI with NVIDIA DGX B200
