Are the Intel Nervana for applications of artificial intelligence
Intel held a presentation on best direction for the industry, creating applications that use artificial intelligence technology (AI). According to Intel, the AI alters the activity of enterprises and the user interaction with the outside world. Intel proposed a variety of technological capabilities in this area cover a wide range of applications, from smart factories and drones, to combat fraud and self-driving cars.
Their development company joined the Intel platform Nervana. According to the manufacturer, Nervana Intel components are designed based on simplicity and ease of use, helping professionals on data processing in different fields to solve the challenges they face.
Because Intel is the largest supplier of processors for servers, which employ AI systems, it is logical that the platform are the Intel Xeon and Intel Xeon Phi accelerators on FPGA. Note that technologies associated with deep learning was acquired by Intel together with the purchase of Nervana, whose name became the name of the platform. Development Nervana pretty soon should appear in the Intel processors. The first test chip called Lake Crest, will be ready in the first half of 2017. Later they will gain key customers. In addition, Intel plans the release of the product under the code name Knights Crest.
The company is confident that their innovations will be able to the end of the decade to achieve 100-fold performance increase in training neural networks.
Among the next stages of the strategy — the release of the Intel Xeon Phi is a new generation, known as the Knights Mill. It is expected that they are four times superior to their predecessors in the speed of deep learning neural networks. These processors should come to market in 2017. Already started the delivery of a preliminary version of the processor Intel Xeon new generation (Skylake) to some cloud provider. The innovation in these processors, is the extension of AVX-512, which also allows to increase the speed of deep learning neural networks.