News

Intel’s AI Portfolio Shows How Designs Must Gear Up for Real-Time Processing

June 22, 2020 by Adrian Gibbons

With processing demands on data centers higher than ever before, designers should be aware of how hardware- (and software-) based AI may have increasing prominence in their designs.

Data center engineers may perk up at Intel's announcement of its next-generation portfolio of Xeon processors, Optane persistent memory, and Stratix FPGA—all with optimized artificial intelligence (AI) integrated onboard.

 

Intel's AI-optimized FPGA

Intel's first AI-optimized FPGA. Image used courtesy of Intel Corporation
 

AI is becoming a mainstay in several sectors, from healthcare to transportation. In fact, Intel claims that 75% of commercial enterprise apps will use AI by 2021. 

Intel's Lisa Spelman (corporate VP of data platforms group/general manager of Xeon and memory group) explains that cloud architecture is being extended to the 5G network and edge devices through the adoption of “compute” heavy devices; AI is fundamental for turning “data from a burden, into an opportunity," she explains.

In that vein, we'll explore how AI—especially at the hardware level—is changing the way we work every day.

 

Highlights of Intel's AI-Focused Portfolio

Intel has set itself apart in the market with this new generation of Xeon technology, providing scalable processors (four and eight sockets), with up to two times the gains over existing infrastructure. Intel says the new processors are the first mainstream server processor including bfloat16 support.

This device is designed for CPU-deployable AI inference and training. Applications can range from image classification and recommendation engines to language modeling and speech recognition. 

 

3rd Generation Intel Xeon Scalable processors

3rd generation Intel Xeon scalable processors. Image used courtesy of Intel Corporation
 

Optane provides much denser memory capabilities (512 GBs vs DRAM 32 GBs) as well as the ability to operate securely and maintain the data without power.

Stratix 10 AI-enabled technology is said to reduce or eliminate connectivity bandwidth issues as a function of multi-node design, which is required for real-time applications like video analytics.

Other highlights in the portfolio include its 3D NAND SSDs, oneAPI cross-architecture tools portfolio, increased IT support.

Intel's strides in AI may represent a shift in responsibilities for circuit designers—one in which they may be required to be fluent in both hardware and software design.

 

How AI May Affect the Electrical Engineer

A few examples across the industry illustrates the power of AI for the modern circuit designer. 

One example comes from researchers from the U.S. Department of Energy’s Lawrence Berkeley National Laboratory, who used a machine-learning algorithm to discover previously-undiscovered connections in old materials science papers. In 2019, team leader Anubhav Jain mused, "Probably the most interesting thing we figured out is, you can use this algorithm [Word2vec] to address gaps in materials research, things that people should study but haven’t studied so far.”

This technology could be incredibly useful in forging new discoveries in electrical engineering, perhaps making breakthroughs in fields like quantum computing. 

Meanwhile, startup JITX is using AI software integration to accelerate PCB design by making hardware more like software, fundamentally changing the role of a PCB designer. Traditionally, PCB designers and drafters will place and route the logical interconnections in order to complete the physical PCB design for manufacturing.

 

JITX

JITX says they can help engineers design circuit boards with their IDE. Image used courtesy of JITX
 

Should companies like JITX become mainstream, the current role of a PCB designer may shift, promoting engineers who have a blended skillset in hardware and software algorithm design with the low-level rote design work being completed by AI “peers."

 

Real-Time Processing in Healthcare, Data Centers, and Even Video Games

We've recently discussed how AI devices have helped researchers stay one step ahead of COVID-19 while also shouldering the burden of increased data processing—a result of stay-at-home orders. Intel claims that this new portfolio will help circuit designers, especially those who work with data center technology, to lead these increased demands as they assist medical professionals and researchers

Engineers in data centers are facing a two-pronged issue: 1) maintaining and upgrading existing networks to meet bandwidth demands and 2) monitoring and responding to malicious activities on their networks.

 

Intel Optane persistent memory 200

Intel Optane persistent memory 200 series. Image used courtesy of Intel Corporation
 

Neural networks can play a role in predictive and preventative maintenance, energy savings, and intelligent security. However, these tasks also put a burden on the existing infrastructure. Intel estimates that by 2025 that a quarter of all data will be generated in real-time, indicating that more power- and memory-efficient devices, like the new Intel portfolio, may be necessary as infrastructure upgrades for better data processing.

This capacity to monitor data in real-time will bring huge changes, ranging from real-time trading on Wall Street to the real-time flight information required by the aviation industry. Blizzard Entertainment President J. Allen Brack has also reported that the company uses AI to monitor and respond to player-driven negativity in their games, including hate speech and racial slurs

 

A Blend of Hardware and Software Skills

For board-level designers, AI will pave new roads in the coming decade, where fundamental hardware knowledge must be paired with a deep understanding of machine learning algorithms and software coding skills. 

Intel’s new portfolio is both a notable shift in performance as well as a call to action. As the demands of modern computing continue to put a strain on existing data centers, the march toward real-time processing will continue.

1 Comment