Microsoft: Project Brainwave

Project Brainwave is the platform intended for acceleration of deep learning of artificial intelligence with a computing power of 39.5 teraflop (or 39.5 trillion operations per second).

Платформа получила четыре модуля FPGA, created by Intel company based on 14-nanometer Stratix 10 processors

2017: Announcement

The Microsoft corporation submitted on August 28, 2017 the Project Brainwave platform for acceleration of deep learning of artificial intelligence. Its computing power is 39.5 TFLOPS. Such high performance and midget waiting time allow AI to process from Microsoft operational data in real time that is crucial during the work with the search queries, a streaming video given from sensors or interaction with users. The platform is capable to execute 130 thousand computing transactions for one cycle, declared in corporation.

The Project Brainwave platform consists of three main layers: a high-performance distributed architecture, the hardware engine of deep neural network (DNN) constructed on FGPA (programmable gate arrays of Field-Programmable Gate Arrays); the compiler with library of time of execution.

The platform received four modules FPGA created by Intel company based on 14-nanometer Stratix 10 processors. FPGA represent programmable hardware which not only increase the speed and productivity of work, but also provide the necessary flexibility. Having involved only one node of structure of FPGA, it is possible to translate all 1440 pages of "War and peace" from Russian into English in two and a half seconds. If to involve all resources of the supercomputer based on artificial intelligence, it is possible to translate completely Wikipedia (which in printed form would be a pile of sheets more than 400 meters high) quicker, than for the one tenth fraction of a second.

At the moment Project Brainwave supports Microsoft Cognitive Toolkit and Google Tensorflow. In plans of corporation — to add support of a set of other platforms and also to implement Project Brainwave in a cloud service of Azure in 2018 that any client could get access to this technology.