Intel readies 10nm chips for AI, 5G and a new PC platform

During CES, Intel revealed how it would address the market opportunity in 5G and support AI inference

Intel used the start of the Consumer Electronics Show (CES) in Las Vegas to present its vision to become a key technology provider in 5G networking and artificial intelligence (AI).

According to Intel’s client computing group senior vice-president, Gregory Bryant, the data-centric market is worth $300bn, and Intel hopes its plans to target data processing, manipulation, advanced analytics and deep learning will be the largest opportunity the company has ever pursued.

“We are at the early stages of a new era in computing, defined by the innovative use of data on every platform from the client to the cloud, from the network to the edge and increasingly everywhere in between,” he said.

On the client side, Bryant discussed an initiative, dubbed Project Athena, to define a new class of PC devices.

He said Intel had taken a new approach to developing systems on chip designs for hybrid computing. During a demo at CES, Intel showed a new client platform, codenamed Lakefield, featuring the first iteration of its Foveros 3D packaging technology.

According to Intel, this hybrid CPU architecture combines different pieces of chip technology that might previously have been discrete components into a single product with a smaller motherboard footprint, enabling hardware providers to create more flexibility in thin and light form factor designs. Lakefield is expected to be in production this year.

Bryant described how it would be possible to configure a processor with five cores in a tiny system on chip, paired with its Foveros 3D packaging technology. This would allow hardware designers to create product designs that were smaller and had longer battery life, he said.

From the server perspective, Intel datacentre group executive vice-president Navin Shenoy described how the company would be focusing on AI and 5G. “In the AI world, there are two workloads – training workloads and inference workloads,” he said. But the inference aspect of AI is the area where Intel has struggled to make a breakthrough, Shenoy admitted.

To address this, Intel introduced what Shenoy described as a new class of AI chip – the Intel Nervana neural processor. “We expect it will deliver industry-leading performance per watt on real production inference workloads,” he said.

Read more from CES 2019

  • Huawei’s Kunpeng 920 processor is geared towards big data and infrastructure applications, marking its foray into Arm-based servers that are set to change the dynamics of the datacentre market.
  • Here, the former Nokia mapping unit now owned by a consortium of German carmakers, is to integrate Amazon Alexa with its navigation and location services.

Due in the second half of 2019, Facebook is among the development partners, said Shenoy. Facebook is looking at how the Nervana chip could support image recognition on its social media platform, he said.

Regarding 5G, this is another area where Intel had no presence, said Shenoy, but Intel regards 5G as the intersection of communications and computing. “5G requires us to expand our capabilities in the datacentre and bring them into the network,” he said.

As an example, he described how low-latency data processing would be possible only if computing power is moved to where the data is being created. Intel has developed an architecture called Snow Ridge, a system on a chip for 5G base stations, which have traditionally used proprietary hardware, said Shenoy. “We are taking all the tech from the server market and fitting it into a small form factor to sit on a cellphone tower to move servers closer to users.”

According to Shenoy, the Snow Ridge architecture would be able to prioritise application network traffic for low latency. Through Snow Ridge, Intel aims to grow its share of the base station market from zero to 40%  by 2022, he said.

Read more on Artificial intelligence, automation and robotics

CIO
Security
Networking
Data Center
Data Management
Close