Facebook unveils data on its artificial intelligence server

    image
    New Facebook Server

    At the end of last week, Facebook representatives announced that the company's specialists had created a device that embodied the idea of ​​artificial intelligence based on the principle of deep learning . It is assumed that the development can be involved in several Facebook applications, reports venturebeat.com . Anyone can familiarize themselves with the design of the invention in the Open Compute Project.

    The servers, code-named Big Sur, are equipped with graphic processors (GPUs) that serve as the main components of the deep learning algorithm. The method involves the formation of artificial neural networks based on large volumes of information - for example, photographs - which subsequently help to analyze new data. Facebook is investing more and more money and efforts in the development of this area, so why shouldn't the company think about creating the appropriate user equipment, as it did with general-purpose servers, virtual storages and network equipment. Moreover, to share the design of the development is no less a good idea.

    During an online conference on publicly released data, Jan Lekun, head of the Facebook’s Artificial Intelligence Lab, explained: “It's like saying“ this is what we use, and this is what we need. Make more efficient equipment, perhaps we will buy it from you. ” Facebook management prudently invited Lekun to work for the company in 2013.

    The deep learning algorithm, in the field of study of which Lekun is rightfully considered one of the outstanding specialists, allows you to recognize speech, images and even imitate the natural process of language processing. Facebook knows firsthand about all of these technologies. For Facebook, this is the main subject of research, as, in fact, for companies such as Google and Microsoft. Earlier, Facebook published information about some programs dedicated to artificial intelligence, now the list of well-known new products has been added to the recently created equipment samples.

    The Big Sur server can include up to eight GPUs, each with a power of up to 300 watts. Facebook developed Big Sur based on Nvidia's Tesla M40 GPU, although the server design also allows for other GPUs.

    During an online conference, Lekun noted that Facebook has installed new servers in data centers both within and outside the United States.

    Big Sur servers undoubtedly turned out to be an order of magnitude higher than all previous technologies that Facebook used to improve the deep learning algorithm.

    Here's what Facebook researchers Kevin Lee and Serkan Piantino write on their blog: “Thanks to NVIDIA’s accelerated Tesla computing platform, Big Sur is twice as fast as its previous generation counterparts, which means we can run the necessary mechanism twice as fast and analyze data received from the network twice as efficiently. And the load distribution between eight GPUs allows us to multiply the results by two more. "

    Also popular now: