NVIDIA DGX Supercomputers

Designed to meet the computing needs of deep learning and data science, NVIDIA DGX™ systems are built on the revolutionary NVIDIA Volta™ GPU architecture. With innovative GPU-optimized software and simplified management, these solutions deliver breakthrough performance and productivity.

NVIDIA DGX systems are designed to give scientists the most powerful tools for AI development on any platform, from the desktop to the data center to the cloud.

Train larger neural network models faster and get results from day one.

Break the limits of AI computing with NVIDIA DGX-2. It’s the first 2-petaflop system with 16 fully interconnected GPUs, delivering up to 10x faster deep learning performance. Powered by NVIDIA® DGX™ software and built on the NVIDIA NVSwitch architecture, DGX-2 is designed to tackle the world’s toughest AI problems.

Unmatched compute power delivers unprecedented training performance

As AI algorithms become more complex, they require unprecedented levels of compute power. NVIDIA® DGX-2 combines the power of 16 of the world’s most advanced GPUs to accelerate new types of AI models that were previously impossible to build. And it’s designed to scale massively. Train up to 4x larger neural network models while experiencing 1x the performance of 8-GPU systems.

NVIDIA DGX-2 is now available in two models, including the new and improved DGX-2H. It is specifically designed to deliver maximum performance for even the most demanding applications. Learn about the DGX-2H, the compute core of the DGX-2 POD. It is the first AI compute infrastructure in the list to be ranked in the Top500 by performance.

The NVIDIA DGX-1 system includes a comprehensive suite of software optimized for deep learning, enabling data scientists and researchers to quickly and easily train deep neural networks. The DGX-1 software suite includes the NVIDIA Deep Learning GPU Training System (DIGITS™), a complete interactive system for building deep neural networks (DNNs), as well as the NVIDIA CUDA® Deep Neural Network (cuDNN) version 5, a GPU-accelerated library of DNN primitives. In addition, the system includes optimized versions of several widely used deep learning frameworks—Caffe, Theano, and Torch. DGX-1 also provides access to cloud-based management tools, software updates, and a bank of container applications.

The NVIDIA® DGX Station™ is the world’s first personal AI supercomputer designed for the office. It runs the same NVIDIA GPU Cloud Deep Learning software stack as other NVIDIA® DGX™ systems, allowing you to work in the office and deploy your neural networks to other DGX systems and the cloud.

With a footprint no larger than a desktop PC, DGX delivers the computing power of four servers at one-twentieth the power consumption. It is the world’s only personal supercomputer equipped with four NVIDIA® Tesla® V100 accelerators with second-generation NVIDIA NVLink™ technology and the new Tensor GPU architecture. DGX trains deep learning algorithms 2x faster than the best workstations. It is equipped with a liquid cooling system and delivers 500 teraflops of half-precision (FP16) operations.