Intel lays out its AI strategy until 2020
No sign of GPUs though
Intel has flexed its AI muscles and beefed up its services with a bunch of new products and collaborations, in an effort to adapt to the technological upheaval of intelligent software.
At Intel’s first “AI Day” in San Francisco, Brian Krzanich, CEO, said the company is “continuing to evolve” and working to provide an “end-to-end AI solution” to allow companies to easily integrate intelligence into their infrastructures.
As data generated by companies continues to pile up, the interest in analyzing that data using machine learning and AI has been piqued. The largest technology companies are all making big investments and staking their claims in AI.
But while companies such as Google and Microsoft have developed libraries of machine learning tools such as TensorFlow and Cognitive Toolkit, Intel is more focused on updating servers to cope with the intense computation required to process and train AI systems.
It has a long history in the semiconductor business, and is making a bold move to retain its "Chipzilla" status by developing its own AI chips. Unlike its main rival Nvidia, Intel has decided to steer clear of GPUs (graphics processing units) – an area where it has little influence – and is instead offering its “Lake Crest” chips, which will be available in 2017.
A more powerful chip code-named “Knights Crest” is also in development to be integrated with its Xeon processor series. Both chips are geared toward powering neural networks for deep learning, and promise to optimize performance.
Diane Bryant, EVP and GM of the Data Center Group at Intel, announced that Intel plans to slash the time it takes to train neural networks 100-fold by 2020 through its chips and processors.
It’s an ambitious goal, and heavily supported by Nervana – a deep learning company that Intel acquired in August.
A preliminary version of Intel’s latest Xeon processor, known as “Skylake,” has begun shipping to select cloud service providers. The next upgrade to its Xeon Phi processors code-named “Knights Mill” will be available in the next year.
With the help of Google, Intel is moving into the cloud space, as both companies announced a partnership to adapt Google’s open-source container cluster system Kubernetes and TensorFlow to Intel’s architecture. Working together will also strengthen security between Intel’s IOT and Google Cloud, Bryant said.
Following suit of other AI companies, Intel also announced it had launched an AI strategy board made up of four researchers: Yoshua Bengio (University of Montreal), Bruno Olshausen (University of California, Berkeley), Jan Rabaey (University of California, Berkeley) and Ron Dror (Stanford University).
To do its part in “democratizing” AI, Intel has partnered with Coursera, an online education website originally set up by Andrew Ng, Chief Scientist at Baidu Research, and Daphne Koller, AI researcher at Stanford University, to bring a range of AI courses to the public.
“Intel can offer crucial technologies to drive the AI revolution, but ultimately we must work together as an industry – and as a society – to achieve the ultimate potential of AI,” said Doug Fisher, SVP and GM of the Software and Services Group at Intel. ®