NVIDIA and Facebook Team Up to Supercharge Caffe2 Deep Learning Framework

by Kristin Bryson

NVIDIA and Facebook today announced the result of our joint work to advance artificial intelligence with Caffe2, a new AI deep learning framework contributed by Facebook to the open-source community.

Every day, the world generates information — text, pictures, videos and more. Facebook is developing new AI systems to help manage this information so people can better understand the world and more effectively communicate, even as the volume of information increases. Caffe2 allows developers and researchers to create large-scale distributed training scenarios and build machine learning applications for edge devices.

Providing AI-powered services on mobile is a complex data processing task that must happen within the blink of an eye. Increasingly, the processing of lightning-fast AI services requires GPU-accelerated computing, such as that offered by Facebook’s Big Basin servers, as well as highly optimized deep learning software that can leverage the full capability of the accelerated hardware.

NVIDIA and Facebook are delivering AI acceleration through our work on the Caffe2 deep learning framework. Thanks to our joint engineering, we’ve fine-tuned Caffe2 from the ground up to take full advantage of the NVIDIA GPU deep learning platform. Caffe2 uses the latest NVIDIA Deep Learning SDK libraries — cuDNN, cuBLAS and NCCL — to deliver high-performance, multi-GPU accelerated training and inference. As a result, users can focus on developing AI-powered applications, knowing that Caffe2 delivers the best performance on their NVIDIA GPU systems.

Caffe2 is designed to be a fast, scalable and portable deep learning framework. It delivers near-linear scaling of deep learning training with 57x throughput acceleration on eight networked Facebook Big Basin AI servers with 64 NVIDIA Tesla P100 GPU accelerators. This means developers can train and iterate AI models faster than ever.

As part of the companies’ collaboration, the NVIDIA DGX-1 AI supercomputer will be the first AI system to offer Caffe2 within the optimized software stack for deep learning. Together, DGX-1 and Caffe2 deliver high performance and fast training. Caffe2 for DGX-1 will be available to customers via NVIDIA DGX-1 Container Registry.

Caffe2 GPU performance chart

Caffe2 Training Through NVIDIA Deep Learning Institute

Through NVIDIA’s Deep Learning Institute, we’ve helped more than 10,000 developers worldwide learn to use frameworks to design, train and deploy neural network-powered machine learning for a variety intelligent applications and services.

Starting at our GPU Technology Conference on May 8-11, we’re adding Caffe2 training to the DLI curriculum. Members of our DLI team will be on hand to help developers get their hands dirty with Caffe2 through a self-paced lab.