Intel plans to make AI '100x faster' with new computing chips

Posted Nov 18, 2016 by James Walker
Intel has unveiled a new series of chips that it says could improve one metric of artificial intelligence performance by up to 100 times today's benchmarks. The next-generation processor range is aimed firmly at deep learning and neural networks.
Intel made the announcements during its inaugural "AI Day" today. It explained how it plans to integrate technology from Nervana, a company it acquired just two months ago, into its wider product range. It expects to see Nervana tech inside Intel silicon in the first half of 2017.
Intel is building two new series of Xeon processors called Lake Crest and Knights Crest. Lake Crest is optimised for neural networks and is designed specifically to offer the highest possible performance for training AI systems. It offers "unprecedented" computing density, according to Intel.
The company expects to deliver a 100x reduction in the time taken to train new complex neural networks by the end of the decade. The breakthrough improvements will be driven by technology developed by Nervana that increases the maximum density of computing architectures. It removes some of the limitations of traditional systems.
"We expect the Intel Nervana platform to produce breakthrough performance and dramatic reductions in the time to train complex neural networks," said Diane Bryant, executive vice president and general manager of Intel's Data Center Group. "Before the end of the decade, Intel will deliver a 100-fold increase in performance that will turbocharge the pace of innovation in the emerging deep learning space."
Intel is also planning a next-generation line of Xeon Phi processors. This range, code-named Knights Mill, will be available in 2016 and will offer up to 4x increased deep learning performance over today's systems. The new chips will give AI developers more flexibility when developing complex systems, enabling AI to do more in real-time. Training new systems will be significantly faster, enabling a new wave of AI that's more advanced.
Intel is treading a very different path in AI to that of its industry rivals. Whereas Intel is continuing to stick with CPU-based technology, companies such as NVIDIA are gaining ground by bringing GPUs to the table. Intel CEO Brian Krzanich dismissed GPU tech as a long-term solution, describing it as an "opportunistic" approach to developing AI. He suggested CPU architectures such as Lake Crest, Knights Crest and Knights Mill will enable the next wave of AI development, pointing to Intel's broad product range as a significant asset in the years to come.
"Some scientists have used GPGPUs because they happen to have parallel processing units for graphics, which are opportunistically applied to deep learning," said Krzanich. "However, GPGPU architecture is not uniquely advantageous for AI, and as AI continues to evolve, both deep learning and machine learning will need highly scalable architectures. Intel architecture can support larger models and offer a consistent architecture from edge to the data center."
Intel also announced a new "strategic alliance" with Google. The two companies will work together to create an "open, flexible and secure" multi-cloud infrastructure that will mutually benefit both businesses.
Separately, Intel will donate $25 million to the Broad Institute to help develop high-performance computing systems for genomics analytic projects. The money will help to create new infrastructure to help scientists integrate and process genomic data.