At its inaugural “AI Day” event in San Francisco today, Intel talked for the first time about what it will do with artificial intelligence (AI) startup Nervana following its acquisition earlier this year.
At least two types of hardware will come out of the deal, according to a statement.
Intel will test “Lake Crest” chips in the first half of 2017 and will provide them to “key customers” later in 2017.
Intel is also working on “Knights Crest” chips, which will blend Intel Xeon with Nervana’s technology. This is designed to handle neural networks — a core component of deep learning — and will also support “a high-bandwidth interconnect.”
The context is that for deep learning, graphics processing units (GPUs) from companies like Nvidia are widely used. Intel is looking to capitalize on the interest in this area, as Apple, Facebook, Google, Microsoft, and other companies come out with more products that use deep learning.
Now Intel is saying that by 2020, Intel infrastructure will enable training of neural networks at 100 times the performance on today’s GPUs, said Diane Bryant, executive vice president and general manager of Intel’s data center group.
Intel is also working on a Nervana Graph Compiler that will be released under an open-source license in early 2017. But the company has also been working on existing deep learning frameworks like TensorFlow, Theano, Caffe, and Torch. Additionally, by the end of this year, Intel will release BigDL software to bring deep learning capability to the Spark data processing engine.
Also today, Intel is announcing the launch of the Intel Nervana AI Academy training program.