As telcos and cloud providers, among others, chase the edge computing opportunity, it is no wonder that the big chip suppliers are sniffing a welcome source of new revenues. Both Intel and Qualcomm placed edge computing at the heart of their presentations at last week’s Consumer Electronics Show (CES) in Las Vegas, setting the scene for yet another head-to-head between the two US rivals.
Intel said its 2020 strategy would be built around three prongs – the intelligent edge, 5G and artificial intelligence (AI), all of which intersect heavily in many applications. For instance, it will use AI to move data around networks to the most appropriate edge location to be processed or stored, using high speed 5G connectivity.
“These technologies are not standalone technologies, they’re actually reinforcing,” said Intel CEO Bob Swan. “5G is enabling intelligent devices, but intelligent devices are putting challenges on developing the technologies for AI and for 5G, so they’re not independent technology inflections. The challenge and our role is to fuel silicon technology and software at the core of each and every one of these technologies.”
The world’s largest chip supplier is banking on a great proliferation of device types – each needing different chip combinations – after years when the smartphone did virtually everything. It is looking for new edge and IoT applications increasingly to require specialist devices, such as industrial wearables, as functionality grows.
Swan predicted that connected devices would generate 175 zettabytes of data by 2025, and 50% of that would be created by devices at the edge, “spanning smart devices, wearables, smart industrial factories, technology being deployed to make retail shopping experiences greater, and connected cars”.
Today, he argued, only 1% of that data from connected devices is being processed and analyzed by AI models, but “AI is making 5G and this convergence of compute with communications to be more and more relevant. These billions of devices at the edge need to be able to have compute closer to them.”
The aim is to embed AI into chips for everything from devices to cloud infrastructure. While the latter is the market where Intel has the biggest in-built advantage, because of its massive dominance of the server space, it has its eyes on a whole AI processor portfolio that reaches from cloud to edge, IoT gateway to device. After all, its lead in servers will be threatened by the webscale giants’ tendency to seek second sources of processors, or to commission their own; and by the determination of China to become a silicon powerhouse, often harnessing the ARM architecture, as Huawei is doing for its own cloud and AI chips.
Intel said its third generation of Xeon Scalable processors, which are coming in the first half of this year, will feature in-built AI training acceleration and up to 60% increase in training performance, said the company.
Meanwhile, Qualcomm’s big CES theme was automotive, but it also had a new product for edge computing and micro-data centers.
“5G is changing the cloud with the appearance of the edge cloud that really gets processing done at the edge,” said Qualcomm president Cristiano Amon, and this can take full advantage of 5G’s bandwidth and latency advances.
The edge is Qualcomm’s latest entry point to the infrastructure market. It has made several attempts from different angles, but never presented any real danger for Intel – while in reverse, Intel’s various bids to be a major player in mobile device processors have also failed, leaving Qualcomm’s dominance challenged not by its compatriot, but by Chinese rivals.
Qualcomm’s most ambitious attack on the data center was Centriq, a 38-core ARM-based processor which it announced in 2017 and was targeting at low power use cases such as Cloud-RAN, blade servers and coprocessors. However, a year later it put the effort on the back burner, leaving only its Chinese joint venture HXT to fly the flag.
Its new effort is more modest but targets the edge data center, where the Intel architecture, and its usual margins, may not be a perfect fit, and where the ecosystem is very immature. Amon showed off an intelligent edge box based on Qualcomm silicon, which is designed to work with the Cloud AI 100 data center card, which the company unveiled last year.
Qualcomm teased the audience with the news that a “major hyperscaler” was already building a 5G edge box using its platform, including the AI 100. Amon said: “Right now you see the Qualcomm logo on this but it’s the result of a design plan with a major hyperscaler company that is actually building the first 5G intelligent edge box for 5G connectivity in AI processing and using AI 100.” He said more details would be shared in the coming months and he expected the box to be “deployed commercially at scale in the second half of this year”.