Bravo
If ARM was an arm, BRN would be its biceps💪!
Intel pushes grand plan to rule the edge, both network and compute, at MWC
By Matt HamblenFeb 26, 2024 3:18pmInteledge AIMWC 2024Generative AI
Intel is trying to rule the edge computing and networking environment with new software and silicon announced Monday at MWC and is positioning these new products as integral to its AI Everywhere strategy outlined last fall.
At the same time, Intel sees itself as the tech provider best prepared to handle AI inferencing, almost as if AI training will become routine, even passe, in coming years. Nvidia has wracked up record revenues by becoming the master of AI training (alongside inferencing) routines with its advanced GPUs used practically everywhere, but Intel’s line of reasoning is that its CPUs are best suited for the inferencing work where the real AI action is and will be.
Intel CEO Pat Gelsinger made this overarching appeal last October and the company has been steadily reinforcing the strategy. On Monday, Intel’s Pallavi Mahajan spoke to Fierce Electronics via phone to carry forward a similar message. She’s the vice president and general manager of Intel’s network and edge business, but the Intel strategy around AI Everywhere surely reaches beyond networking, and even beyond what CIOs might define as the edge. Edge environments become more complex with security and privacy demands.
“AI is a pervasive workload,” she said. “When we think of AI and OpenAI, where we’re training big models,…those models will soon get standardized and commoditized. More of the computing will be with inferencing with CPU affinity, where Intel is leading. The use case starts on my laptop, and in stores with self-checkout. Intel is very focused on AI Everywhere, not just training big models, but how to make inferencing happen every day. I feel very bullish about how we’re doing it.”
At MWC, Intel on Monday announced a future Xeon processor with AI acceleration, code-named Granite Rapids-D for 5G vRAN and also previewed its next-gen Xeon for 5G core, code-named Sierra Forest. The third announcement was Intel Edge Platform, software to help enterprises build, deploy run and manage edge and AI systems on standard hardware. The company said it will also announce a new Intel vPro platform on Tuesday to extending AI PC capabilities to commercial designs.
Sierra Forest will launch later this year to offer 2.7x performance per rack improvement. Intel said BT Group, Dell, Ericsson, HPE, KDDI Lenovo and SK Telecom have shown interest in the platform. Granite Rapids-D is planned to launch in 2025, following the launch of Granite Rapids server CPUs in 2025. But Granite Rapids-D is sampling and Samsung has demonstrated a first call at the R&D lab, while Ericsson has demonstrated it at a joint lab with Intel in Santa Clara. Intel is also working with Dell, HPE, Lenovo, Mavenir, Red Hat and Wind River on the product, Intel said in a release.
An Intel vRAN AI development kit is also going out to some Intel partners such as AT&T, Deutsche Telekom, SK Telecom and Vodafone to show what AI can bring to RAN.
Mahajan said Intel has been working with 90,000 edge customers globally for the past decade, selling more than 200 million processors at the edge.
“Every enterprise knows edge is growing, and it’s very heterogeneous, with different generations of hardware and software,” she said. “It’s a problem of scale. If you are a quick serve restaurant, every store today you have needs a node.” With a modern software-defined approach, preferably Intel’s approach, every time a restaurant needs a security patch, it won’t have to send someone to the restaurant to do the patch, she said.
Already on the first full day of MWC, she said she had met with customers who want to run AI on their existing devices at the edge, which is where Intel Edge Platform software makes a difference. “It brings in a single data class for management of your infrastructure, which means its simple to build apps in a local development application environment,” she said. OpenVino helps allow AI inferencing real time.
For a retail store, engineers want automated checkout to happen, but they might want to add other features where inventory management and traffic analysis of customer visits happens on the same server, she said.
One visitor to the Intel booth at MWC, she said, was the Ukraine military service which wanted help with creating a smart power grid that can be kept secure from power grid cyber attacks. Another visitor, a factory manager, was interested in using the edge platform to automate a factory to avoid cyberattacks.
She said Lenovo has been using current Intel silicon with the edge platform on highly ruggedized wall-mounted servers for a global fast-food chain. But it’s not just retail, she said. “We’re focused on every vertical and most of them have some common use cases.”
Intel named a number of Intel partners demonstrating at MWC: Aira Technologies, ADLINK, Corning, DeepSig, Ericsson, Federated Wireless, Haivision, Hitek Systems, Lanner, Mavenie, NEC, Nokia, Samsung, Senao Networks and Tiami Networks. On its website, Intel also included an explainer of its Edge Platform.
Last edited:
what’s up Australia
