Intel unveiled 10-nanometer (nm) chips for data center servers and 5G wireless access base stations and said it’s teaming up with Facebook on new artificial intelligence (AI) chips at a CES 2019 news conference on Monday.
The chip giant said Facebook is one of its development partners on the new Nervana Neural Network Processor for Inference. It’s a new class of chip designed for AI inference workloads — this is when a machine acts on a new data sample to infer an answer to a query. Intel expects the new processors to go into production in the second half of this year.
“This is a really big deal for us,” said Intel Data Center Group Executive Vice President Navin Shenoy. “It expands our position in AI above and beyond what we’ve done in Xeon and in Core into a new domain.”
Most processing advances target AI training. This involves an algorithm’s ability to better understand a data set. “One area where we really haven’t had a breakthrough yet is inference, and this tends to be a more difficult problem,” said Shenoy.
Intel is also working on a chip that tackles the training phase of AI (as opposed to inference). It plans to have a Neural Network Processor for Training, code-named “Spring Crest,” available later this year.
Data Center Chips
Intel also started shipping its latest Xeon Scalable processors, code-named “Cascade Lake,” first announced in November. Cascade Lake includes 48 cores, which improves server performance for demanding workloads, and it also uses Intel Optane DC persistent memory and DL Boost, which is designed to accelerate AI deep learning inference.
Shenoy called Optane DC persistent memory, which Intel has been working on for a decade, “truly groundbreaking.”
“It’s faster than solid state drives, but applications can recognize it as a large memory,” Shenoy said. It provides “persistent memory,” meaning that “data will maintain even when the power is out,” he explained. “Cascade Lake and Optane persistent memory have been designed for data from the ground up.”
And as for its long-awaited 10nm data center chips, Shenoy said Intel Ice Lake server silicon will be available in 2020. He didn’t provide any additional details, but the company previously admitted difficulty developing its processor technology for 10nm chip manufacturing.
Meanwhile, competitors including Samsung and Taiwan Semiconductor Manufacturing Co. (TSMC) have already started mass production of their 7nm chips, which compete with Intel’s 10nm silicon.
10nm for 5G
Despite the delays in data center 10nm chips, Intel said it’s expanding into 5G networking with a new 10nm-based network system on chip (SoC), code-named “Snow Ridge,” designed for 5G wireless access and edge computing.
“The introduction of Snow Ridge is going to bring Intel architecture into a new place where it has never been before,” Shenoy said.
This new place is the wireless access base station. And the new SoC will allow more computing functions to be distributed out at the edge of the network.
“We’re taking all the technology we’ve developed for the server market, we’re shrinking it down, we’re repurposing it, and we’re fitting it into a very small form factor,” Shenoy said.
Snow Ridge is expected to be available in the second half of this year.