Intel revealed new AI chips to compete with Nvidia's GPUs

Details are not much, but Intel says new chips will improve deep learning training time.

When the AI ​​exploded, Intel did not respond. Now they are trying to reaffirm their position in the silicon world by revealing a new line of chips, designed specifically for artificial intelligence, called Intel Nervana Neural Network Processor or NNP for short.

The NNP chip line is like a response to the need for machine leraning and its destination is data centers, not computers.

Intel's CPU may be a giant in the server sector (Intel estimates, it accounts for 96% of the data center market share), but the workload of existing AIs is still due to the microprocessors. graphics or GPU comes from companies like Nvidia or ARM. As a result, the demand for these companies' chips is increasing. (Nvidia's revenue last year increased by 56% next year).

Google has joined, designing its own Tensor Processing Unit chip to support cloud computing, while new companies like Graphcore in the UK have also tried to bridge the gap.

Intel's reaction is to buy AI hardware talents. They bought Mobileye vision company in March, chip maker Movidius (DJI's self-driving chip maker), in September last year and started learning deep learning Nervana Systems in August. / 2016.

Intel invests heavily in AI chip hardware experts

Since then, they are still busy with the Neural Network Processor, formerly known as Lake Crest. NNP chips are the direct result of Nervana acquisition and focus on professionals to achieve 'faster learning time for deep learning models'. (Intel also said to have received a Facebook recommendation for chip design but did not elaborate. '

Exactly how much faster their deep learning model will be, Intel doesn't say. While Google promotes its new TPU chip by offering comparative tests with rivals, https://cloud.google.com/blog/big-data/2017/05/an-in-depth- look-at-googles-first-tensor-processing-unit-tpu Intel just said that they are following the progress and will achieve the goal of speeding up deep learning 100 times by 2020.

Still very vague about when the NNP chip will come to the user.