Facebook Open Sources New AI Hardware

Facebook researchers have open-sourced its latest AI server designs to developers in the open source community. Researchers have developed the hardware capable of training neural networks.

Designed for artificial intelligence at a much larger scale than before, the project was part of a collaboration with Nvidia and is compatible with Open Rack. The Menlo Park-based company adds this project to its Open Compute Project that started in 2011 to offer other organizations a detailed blueprint to set an AI-based architecture.

“There is a network effect. The platform becomes better as more people use it. The more people that rally to a particular platform or standard, the better it becomes—the more people contribute,” says Yann LeCun, who leads Facebook’s Artificial Intelligence Research lab since 2013.

Named Big Sur, the hardware contains GPUs (graphical processing units) as processing chips. Each Big Sur server has 8 GPUs, each operating at 300 Watts. It is designed to fit Nvidia Tesla M40 GPU, however, is capable to fitting other GPUs as well. Originally, GPUs were made to render games and other 3D graphical applications, however, their power comes to good use in the area of artificial intelligence. Facebook has accommodated Big Sur servers in data centers all over the world.

Big Sur focuses on Deep Learning – a type of artificial intelligence that helps with identification of patterns in large pools of data. It works like a human brain, mimicking the neural networks and the path they take to identify patterns. Deep learning can be applied to image recognition, speech recognition and NLP (natural language processing).

This move is one of many altruistic projects undertaken by Facebook. “This is a way of saying, ‘Look, here is what we use, here is what we need. If you make hardware better than this, we’ll probably buy it from you,'” Yann LeCun stated.

Just last month, Google released TensorFlow, its "open source software library for numerical computation using data flow graphs. Nodes in the graph represent mathematical operations, while the graph edges represent the multidimensional data arrays (tensors) communicated between them. The flexible architecture allows you to deploy computation to one or more CPUs or GPUs (Graphical Processing Units) in a desktop, server, or mobile device with a single API." However, contrary to Big Sur, Google did not release hardware to support TensorFlow.

Microsoft is also in the AI machine learning race, along with Google and Facebook. The technology would assist all the companies in their operations.