Intel has introduced a new family of chips designed especially for artificial intelligence (AI). These are the Nervana Neural Network Processor (NNP) processors, which arise as a response to the needs of automatic learning.
The CEO of Intel announced that he is very excited to have Facebook in close collaboration by sharing its technical knowledge as new generations of hardware are added to the market.
This announcement was made during the inauguration of the annual WSJD Live conference.
Unlike its computer processors, the new Intel Nervana NNP chips have an Application Specific Integrated Circuit (ASIC) specially designed for the training and execution of deep learning algorithms. The speed and computational efficiency of deep learning can be very advanced by ASICs that are customized for this workload.
One of the biggest news is that Intel has eliminated the standard cache normally seen on CPUs, which has been replaced by special software that manages processor memory with a given algorithm. This allows the chip to reach new levels of computation density and performance for deep learning.
It is also designed with high-speed interconnects, both inside and outside of the processor, allowing you to perform massive bi-directional data transfers. That means if you put a lot of these chips together, they can act together like a big virtual chip, allowing deeper and bigger learning models.
However, Intel used a smaller form of math, called Flexpoint. Neural networks are very tolerant of ‘noise’ data and this noise can even help them converge into solutions. At the same time, using smaller precision numbers allowed designers to increase parallelism, reducing latency and increasing bandwidth.
The new design of Intel products has as its main goal to develop a processor that can be able to operate all the workloads of deep learning and efficient enough to be able to handle and operate with the best features of high-intensity computing and make the components are becoming more optimal. By applying artificial technology to Intel processors, they can develop each task better and faster. In addition, it is possible for each processor core to work better and several tasks can be performed simultaneously.
Last August, Intel bought $ 350 million for Nervana Systems, a deep-learning hardware startup whose technology is tightly linked to these new processors. Intel Nervana NNP chips will be available later this year and it looks like the company plans to launch an entire line of products around these new processors. With the new systems Nervana Intel is innovating in its products using artificial intelligence to improve its processors more and more.