MIT engineers have developed a powerful processor for artificial intelligence

Not far off is the day when the rebels against humanity send the cyborg-killer to the past, because a team of specialists from the Massachusetts Institute of Technology developed a a processor that significantly speeds up the process of data processing and image recognition by neural networks. Their development in the relatively near future can give wearable electronics the ability to think and make decisions.

The processor, codenamed Eyeriss, was presented at the international conference of microelectronics manufacturers in San Francisco. Eyeriss was originally developed to facilitate the process of in-depth training. The creators of the processor claim that their chip is capable of 10 times higher in the processing speed of some modern GPUs. In other words, this technology will very soon be adaptable to portable devices like tablets and smartphones.

"The application of the process of in-depth training is very important in a variety of areas. This set of algorithms allows for the recognition of objects, speech, human faces. Today's neural networks are very complex and able to work exclusively on powerful GPUs. Just imagine that you will be able to get all this functionality on your smartphone, "- says the assistant professor of electrical engineering at MIT Vivienne.

An impressive increase in data processing speed was achieved due to a very cunning design of the processor. First, instead of using a common memory pool, each of the 168 Eyeriss cores has its own discrete cache. This allowed to get rid of excessive data transfer within the system. Secondly, in order to reduce the data transfer rate, the processor first compresses the information using a certain contour, and then decompresses it at its destination.

But most importantly, the Eyeriss processor uses a special ingenious scheme, through which the data processing process can be configured so that the data between the cores are distributed in the most efficient way. Evenly distributing the load, the processor per unit of time processes much more data than some modern GPUs. However, so far nothing has been said about the specific application of the new chip at the conference. It remains only to wait patiently for new information.

The article is based on materials https://hi-news.ru/technology/inzhenery-mit-razrabotali-moshhnyj-processor-dlya-iskusstvennogo-intellekta.html.