This page was machine-translated and may differ from the original. View original
Nvidia has announced the results of its collaborative efforts with Facebook to advance artificial intelligence, working with Caffe2, an open-source AI deep learning framework released by Facebook.
Facebook is developing artificial intelligence systems that help manage vast amounts of data. These AI systems facilitate efficient communication. Developers and researchers can leverage Caffe2 to configure large-scale distributed training scenarios and develop machine learning applications for user devices.
To deliver AI-based services on mobile platforms, complex data processing tasks must be executed instantaneously. Such rapid AI service processing requires deep learning software that can leverage GPU-accelerated computing and fully utilize the performance of accelerated hardware.
Nvidia and Facebook are providing AI acceleration through collaborative work on the Caffe2 deep learning framework. Through joint engineering, end-to-end optimization has been implemented to maximize the advantages of Nvidia's GPU deep learning platform. Caffe2 leverages Nvidia's latest deep learning SDK libraries—cuDNN, cuBLAS, and NCCL—to deliver high-performance multi-GPU accelerated training and inference. As a result, users can recognize that Caffe2 delivers optimal performance on Nvidia GPU systems and focus on developing AI-based applications.
As part of this collaboration, the Nvidia DGX-1 AI supercomputer is expected to be the first AI system to offer Caffe2 in a software stack optimized for deep learning. Both DGX-1 and Caffe2 deliver high performance and fast training. In the future, customers will be able to access Caffe2 for DGX-1 through the Nvidia DGX-1 Container Registry.
To date, Nvidia has supported more than 10,000 developers worldwide through the Nvidia Deep Learning Institute, enabling them to design, train, and build neural network-based machine learning for various intelligent applications and services using the framework.
Caffe2 training will be available through the Nvidia Deep Learning Institute at the GPU Technology Conference being held in San Jose, San Francisco from May 8-11 local time.
To request a correction, reply or follow-up report on this article, see how to file a request. Previously published statements are collected in corrections & replies.
김자영 Reporter















