At MIT developed a powerful processor for artificial intelligence

In the field of high technology, humanity is getting closer to the moment when will be able to send in the past the terminator, if computer engineering will have the opportunity to conduct their own analysis to consistently think in a logical form and on their basis to make a concrete decision, without human intervention. One step in this direction was the latest development of American specialists from Massachusetts Institute of technology.

The scientists were able to create a processor Eyeriss with enhanced capabilities for data processing and recognition of images using neutron network. The development has already demonstrated during the specialized conference in San Francisco.

The authors of the project note that initially set ourselves the task to create a chip that focuses on facilitating deep learning. The resulting product on the background of a modern GPU delivers a tenfold advantage in speed of information processing. In such conditions can be expected the advent of the Eyeriss on mobile devices.

Lately deep learning is becoming more common. This system is constructed from the group of algorithms that allow computer technology to recognize voice commands, the structural features of human faces and various objects. Currently neutron network are already used, but they require the presence of a GPU with a large capacity. The new processor is theoretically enables the use of this technology in these mobile devices like smartphone and tablet.

A significant increase in speed, the engineers managed to achieve through the use of the original design ideas. First, all core, and there are 168 pieces individually selectable caching, which reduced the amount of movement of the information inside the chip. Secondly, before you transfer all of the information is compressed, and after receiving its target again unpacked.

In addition, the work system is built in such a way that the data is distributed between the cores for processing in order to maximize effectiveness and productivity. Due to the uniformity of the loading significantly increases the amount of data processed per unit time.

It should be noted that under the chip makers have not given any specific information regarding the application in practice. It is assumed that new information will be held in the near future.

If you found an error, please highlight the text and press Ctrl+Enter.