MWC 2024 Conference: AI is everywhere, from networks to terminals.

LB Select
2024.02.28 02:30
portai
I'm PortAI, I can summarize articles.

Major chip manufacturers are focusing on 5G and generative AI.

On February 26, 2024, local time, the 2024 Mobile World Congress (MWC 2024) kicked off in Barcelona, Spain. As one of the largest technology exhibitions in the global mobile communications field, MWC 2024, themed "Future First," will focus on in-depth discussions and exchanges on six major themes: "Beyond 5G, Internet of Things, AI Humanization, Digital Smart Manufacturing, Rule Disruption, Digital Genes."

Whenever there is a major international electronics event, it becomes a core battlefield for chip giants to show off their strengths. The ongoing MWC 2024 in Barcelona is no exception.

At this world mobile communication event, chip giants such as Qualcomm, MediaTek, UNISOC, Intel, AMD, NVIDIA, Arm, and others have made a strong presence.

The keywords are very focused: 5G, Wi-Fi 7, Generative AI.

It is worth mentioning that Amazon Web Services, Ericsson, Microsoft, Nokia, Arm, NVIDIA, Samsung, and other major players in the AI and wireless industries have jointly established the AI-RAN Alliance.

The purpose of this alliance is to use AI technology to boost the development of wireless access network (RAN) technology and mobile networks, ultimately improving mobile network efficiency, reducing power consumption, and transforming existing infrastructure.

This trend in the communication network industry has also become a key focus for many mobile chip companies this year.

1. A surge of new 5G chip products, Wi-Fi 7 mobile connection systems supporting AI optimization for the first time

The communication chip products released during MWC 2024 mainly focus on faster 5G and Wi-Fi 7 connection speeds.

In the field of mobile communications, one of the most significant releases is Qualcomm's Snapdragon X80 5G modem and RF system - the first fully integrated NB-NTN satellite communication modem that supports terminal connections to non-terrestrial networks. In addition to 5G Advanced features, it also features a dedicated tensor accelerator to support AI optimization.

MediaTek unveiled the 5G RedCap (lightweight 5G) product lineup's new member, the MediaTek T300 platform, which integrates the RF system and features the MediaTek M60 modem, providing high connection reliability and longer battery life for 5G devices, while reducing product development cycles and costs. The platform has completed 5G SA network connections, VoNR calls, and data transmission tests, suitable for large-scale deployments in IoT, industrial IoT, mobile networking, security, logistics, and other fields.

In addition to announcing the commercial progress of the UNISOC 5G RedCap IoT chip platform V517, UNISOC also showcased the first IoT-NTN satellite communication SoC V8821 chip and terminal prototype at MWC 2024, and launched the industry's first chip platform V620 that fully supports 5G R16 broadband IoT features. V620 is based on the second generation 5G communication technology platform of UNISOC, with a 4-core Arm Cortex-A55 CPU, supporting 5/4/3/2G full network and 5G TSN, with a reserved memory of 100MB+ resources, and can be widely used in various forms of devices such as 5G FWA, 5G handheld terminals, 5G modules, laptops, gateways, and more.

For 5G infrastructure, Qualcomm showcased three AI-assisted network management enhancement features, including a generative AI assistant to help RAN engineers simplify network and slicing management tasks, AI-assisted open RAN applications (rApp) to reduce network energy consumption, and an AI-assisted 5G network slicing lifecycle management suite.

Intel has released two Xeon processors: 1) Sierra Forest, which integrates up to 288 energy-efficient cores on a single chip, expected to be released later this year, improving 5G core network performance while reducing energy consumption; 2) Granite Rapids-D processor, utilizing optimized Intel AVX instruction sets to significantly enhance vRAN performance, integrating Intel vRAN Boost acceleration, currently undergoing sample testing and set to be released in 2025.

Additionally, Intel announced that its Intel Infrastructure Power Manager software aimed at reducing costs and improving efficiency for 5G core networks has been widely adopted by the industry; and introduced a new edge platform for policy-based, unmanned infrastructure, application, and AI management across a range of edge nodes through a single control panel.

During MWC 2024, AMD also showcased advanced technologies and solutions in the fields of 5G, 6G, vRAN, and Open RAN, demonstrating the use of AMD chips for wireless radios and vRAN, low-PHY functions of 8T8R 400MHz radios, and parallel operation of ResNet-based image classifiers. It was mentioned that Ericsson and Telstra are using the fourth-generation AMD EPYC processors to provide energy efficiency and modernization for 5G core network functions; Parallel Wireless deployed GreenRAN 5G DU software on the AMD EPYC 8004 series processors.

Established in 2021, Chengdu communication chip company New Hexton released two 5G lightweight chips, IM6501 and IM2501, targeting entry-level 5G smartphones and the 5G IoT market, both supporting 4G/5G dual-mode. Viettel, a telecommunications company in Vietnam, has announced the launch of a 5G DFE chip. This chip is claimed to have a computing capability of 1 trillion calculations per second, meeting the 3GPP 5G standards and comparable to the 5G DFE chips of the top 10 semiconductor companies globally. Viettel has also introduced an automated network optimization system to improve operational efficiency, reduce energy consumption, and automatically address issues in 5G and 4G networks.

Another highlight of Qualcomm's new product release is Wi-Fi 7: the FastConnect 7900 mobile connectivity system, designed for multi-device interconnection, is the industry's first solution that supports AI-optimized performance and integrates Wi-Fi 7, Bluetooth, and ultra-wideband technology in a single chip. It integrates ultra-wideband technology, Wi-Fi ranging, and Bluetooth channel detection to create a set of short-range sensing technologies.

Furthermore, targeting in-car scenarios, Qualcomm has introduced the industry's first automotive-grade Wi-Fi 7 access point solution, the Qualcomm QCA6797AQ, which can enhance link reliability, reduce latency, increase network capacity, support faster connections, and reduce stuttering or disconnection caused by external interference.

MediaTek also showcased 5G-A satellite communication-related technology solutions and used a robotic arm to demonstrate the process of satellite communication, attracting many visitors to stop by.

The second major focus is on mobile and PC chipsets integrating generative AI, lowering the threshold for deploying large models at the edge.

Generative AI is another highlight of the chip giants' exhibits. Mobile chip giants like Qualcomm and MediaTek have showcased various AI and GC functions based on their own chips, from AI image recognition, AI voice chat, AI image generation, to AI photo editing. UNISOC also exhibited dozens of AI technology applications and features.

Qualcomm demonstrated running multimodal large models with over 7 billion parameters on Android smartphones and Windows PCs, as well as showcasing traditional AI and generative AI functions supported by the Snapdragon digital chassis platform in automotive scenarios.

It is worth mentioning that one of the recently popular AI hardware representatives, Humane AI Pin, appeared at Qualcomm's booth.

MediaTek focused its demonstration on the generative AI capabilities of its Dimensity 9300 and Dimensity 8300 mobile processors, including running text-to-image, video generation, real-time processing of recorded human images, and other edge AI technologies, and demonstrated real-time AI video generation applications based on the Dimensity 9300 mobile chip.

UNISOC also showcased its latest achievements in the AI field, such as the latest UNISOC AI computing platform adopted by the high-performance SoC T820, supporting low-bit quantization of model weight values and activation values, access to mainstream training frameworks in the industry, and advanced weight compression technology, facilitating the deployment of large models at the edge. 紫光展锐 has launched the first AI+8K ultra-high-definition intelligent display chip platform M6780, which also supports complex AI operations.

NVIDIA has introduced two new entry-level consumer graphics cards, RTX 500 and RTX 1000, for laptops. When running models like Stable Diffusion, they can provide up to 14 times the performance of a CPU, 3 times faster AI photo editing, and 10 times better 3D rendering graphics performance. These two GPUs will be launched this spring.

Qualcomm has also provided developers with an AI model library containing 75 pre-optimized mainstream AI and generative AI models, all optimized for edge AI, including Whisper, ControlNet, Stable Diffusion, Baichuan-7B, and more. These models are available on Qualcomm AI Hub, Hugging Face, and GitHub, leveraging the hardware acceleration capabilities of all cores in the Qualcomm AI engine to increase inference speed by 4 times.

Conclusion: From the network to the end device, AI is everywhere.

Overall, at MWC 2024, major chip manufacturers have released more new products, showcasing how their technologies, products, and solutions are further enhancing connection speed, efficiency, and energy performance in areas such as networks, cloud, edge computing, PCs, and smartphones.

AI technology greatly benefits optimizing application experiences in these areas. From infrastructure to hardware, with the support of these chip companies, AI is penetrating into more specific scenarios and industries, helping address increasingly complex enterprise workloads challenges and continuously improving consumer experiences on end devices.