As you know, artificial intelligence requires huge computing power. Nowadays, the most common components are often used as electronic components (with the exception of the most powerful systems like IBM Watson). Typically, Intel solutions from consumer Core to Xeon are used. But they were all created for slightly different tasks that do not involve working with neural networks. Therefore, the company's management decided to develop a processor that would be sharpened specifically to work with AI systems.
Intel's first dedicated AI processor is dubbed the Nervana Neural Processor (NNP), with the first models slated to ship later this year. The new Nervana processor is an application-specific integrated circuit (ASIC) chip, and is primarily aimed at performing tasks in the field of deep machine learning and self-learning. The processor is based on data buses that provide bidirectional exchange of large amounts of information. Thus, it will be possible to combine multiple Nervana processors into one with virtually no loss of efficiency. This approach enables the processor to work with multilevel neural networks. In addition, Intel officials said
“The main feature of the Nervana NNP processor is that it is based on Flexpoint low-precision integer math. The advantage of using it is the increase in performance when performing tasks, decrease in latency and increase the speed of data transfer over external buses. In addition, Nervana can perform matrix addition, intersection finding and a number of other operations at the hardware level. And special algorithms allow you to dynamically change the processor's memory so that its architecture becomes optimal for a specific operation."
Despite Intel's vast experience in the field of creating processors, when entering the market for chips for AI, it will have to face such giants as Qualcomm, IBM and NVIDIA, which have been manufacturing components in this area for quite some time. So it will be very interesting to watch this race.
Vladimir Kuznetsov