Intel Committed to Driving vRAN Innovation with New AI Development Kit and Next-Generation Intel Xeon Processors

Intel Committed to Driving vRAN Innovation with New AI Development Kit and Next-Generation Intel Xeon Processors

● Experimental tests with AT&T, Deutsche Telekom, SK Telecom, and Vodafone show that AI has the potential to improve network energy efficiency and service quality.

● Joint efforts with companies such as Aira, Capgemini, DeepSig, Ericsson, Mavenir, Samsung and Tech Mahindra demonstrate the determination of the broad industry ecosystem to advance the development of AI in wireless networks.

Artificial intelligence (AI) is one of the most promising frontiers in the technology industry, and the mobile industry has already started its AI journey, which includes exploring the benefits of AI in a variety of application scenarios. With the large-scale deployment of vRAN at this stage and the continued steady growth momentum in the next few years, the mobile industry is expected to gradually use the flexibility of vRAN to integrate intelligent functions into RAN. Recently, Intel has achieved some exciting milestones in the mobile industry.

Last summer, Intel released the 4th Gen Intel® Xeon® Scalable processor with integrated vRAN Boost (codenamed Sapphire Rapids EE), which can directly handle acceleration tasks at the physical layer without the need for additional external devices or components. Last year, Intel and Verizon also completed the industry's first data session based on Samsung's vRAN solution. At the same time, Intel also built a solution for Ericsson Cloud RAN for the first time. In addition, Vodafone announced a large-scale deployment of Open RAN in the UK with Samsung, and pointed out that the performance of early Open RAN deployments based on 3rd Gen Intel® Xeon® Scalable processors has exceeded its traditional equipment. Moreover, Vodafone will deploy 4th Gen Intel® Xeon® Scalable processors with integrated vRAN Boost in the UK from April 2024. In addition, AT&T and Ericsson also announced plans to work with Intel to deploy and expand Open RAN services. Recently, Telus and Samsung announced plans to deploy Canada's first commercial virtualized, Open RAN network based on 4th Gen Intel® Xeon® Scalable processors.

These milestones not only highlight the mobile industry’s long-term investment in advancing vRAN and Open RAN, but also demonstrate Intel’s continued commitment to supporting the industry with its leading product roadmap. The next-generation Xeon processor, codenamed Granite Rapids–D, will be released in 2025 and will leverage optimized Intel AVX instructions to achieve significant vRAN performance improvements, integrate Intel vRAN Boost acceleration, and other enhanced architectures and features. The chip is currently being sampled. Samsung has demonstrated the first call at its R&D lab in Suwon, South Korea, and Ericsson has demonstrated the first call verification at the Ericsson-Intel Joint Lab in Santa Clara, California. These important milestones not only highlight the ease of software migration between generations, but also indicate the full readiness of the ecosystem at product launch. To ensure market readiness, Intel has been working with industry-leading ecosystem partners such as Dell Technologies, Lenovo, Red Hat, and Wind River.

Additionally, as we continue to drive Intel’s product roadmap, we are working with the mobile ecosystem to drive RAN innovation through AI, helping operators to fully tap and maximize the value of their existing general-purpose hardware based on Intel Xeon processors, such as adding built-in AI accelerators.

We believe that AI will play a vital role in the RAN environment, helping operators optimize performance, improve energy efficiency and achieve intelligent resource management. Through advanced machine learning algorithms, AI can analyze relevant data generated by the network, identify patterns and predict trends, so as to make real-time decisions to achieve the goals of optimizing wireless resource allocation, improving energy efficiency, and improving user experience and overall network performance. The software flexibility of virtualized RAN enables RAN to continue to evolve, which is also due to the application of AI-based innovations in RAN deployment systems.

We found that compared with computer vision, many RAN AI models are not as deep and extensive, but they have more stringent latency requirements. At the same time, considering the inherent power consumption limitations of RAN, running efficient AI RAN workloads and other RAN stack workloads on the same processor has become a necessary requirement. Therefore, considering the above factors, how to efficiently run AI RAN workloads on processors has become a difficult problem that needs to be solved urgently.

Intel Xeon processors with built-in AI accelerators can run RAN inference workloads within the CPU. This integrated AI acceleration capability makes it possible to deploy optimized AI models on hardware that already has RAN intelligent controllers (RIC), central units (CU) or distributed units (DU), and achieve excellent results.

To help operators take full advantage of AI opportunities based on general-purpose processors, Intel announced that it will provide the Intel vRAN AI Development Kit to some partners in advance. Through this development kit, independent software developers (ISVs), telecommunications equipment manufacturers (TEMs), system integrators (SIs) and network operators can quickly build intelligent functions in RAN without having to invest too much in learning and developing AI expertise.

The development kit includes a set of AI models optimized for vRAN user cases. The development of these models is not only based on optimized data analysis libraries such as oneAPI and frameworks including TensorFlow and PyTorch, but also uses tools such as OpenVINO. In the process of AI development, such commonly used libraries, frameworks, and tools can efficiently generate code optimized for Intel Xeon processors. Based on these optimized AI models, and thanks to the built-in AI acceleration, telemetry, and power management capabilities of Intel Xeon Scalable processors, operators can integrate "intelligence" into the network and realize dynamic configuration of the network to significantly save costs, thereby maximizing the value of infrastructure investment and expanding revenue sources.

In addition, the Intel® vRAN AI Development Kit also includes training code to help ecosystem partners customize their AI solutions based on specific network scenarios. The end-to-end reference architecture included in the kit shows how to integrate AI models with other RAN applications and components such as Intel® FlexRAN Reference Software and RAN Intelligent Controller (RIC).

The AI ​​model inference code in the development kit can not only be deployed anywhere in the network according to the specific architecture of the network operator, but also can be deployed in the cloud as a virtual network function. It is optimized for the performance of Intel Xeon Scalable Processors and can be migrated on multiple generations of Xeon platforms. This compatibility and scalability ensures that software developed for current Xeon processors can be easily run on the next generation of processors with simple recompilation.

The birth of the Intel vRAN AI Development Kit is inseparable from Intel's years of AI technology research and development and close cooperation with a wide range of ecosystem partners. In early versions, the kit includes AI models for key RAN use cases, such as energy saving management, traffic control, and network slicing radio resource management (NSRRM). In the future, we will continue to add AI models for other use cases.

To demonstrate the future of AI-enhanced wireless networks to the industry, Intel is working with AT&T, Deutsche Telekom, SK Telecom, and Vodafone to demonstrate the benefits of AI for vRAN. When testing with operators using the Intel vRAN AI Development Kit, new vRAN processors, and Intel FlexRAN reference software, we have seen many excellent results, including improved energy efficiency, enhanced user experience, and meeting customer service level agreements (SLAs).

At the same time, industry feedback is very encouraging:

AT&T

“Energy efficiency is a high priority for AT&T and the broader telecommunications industry, and AI has the potential to significantly improve energy efficiency,” said Adam Loddeke, vice president of RAN technologies at AT&T. “Our joint testing with Intel has shown that AI-assisted CPU frequency scaling on the FlexRAN distributed unit (DU) can reduce server energy consumption by 16 percent when deployed on a 20-core 4th Gen Intel Xeon Scalable processor with vRAN Boost, which alone can translate into significant cost savings of up to $4.5 million per year in energy savings across a network of 100,000 DU servers. We are excited to collaborate with Intel on AI innovation and plan to deploy 4th Gen Intel Xeon processors with vRAN Boost beginning in 2024.”

Deutsche Telekom Network Test and Integration Laboratory

“We are constantly exploring new technologies and solutions to better provide our customers with an exceptional user experience,” said Petr Ledl, Vice President of Network Trials and Integration Labs at Deutsche Telekom and Chief Architect of Access Disaggregation. “It is still early days for AI in the RAN, but our testing collaboration with Intel shows that AI on x86 computing platforms combined with O-RAN architecture can effectively improve vO-DU beam selection at the edge of the cellular network. We are pleased that Intel is helping to jointly advance this emerging technology based on general-purpose processors.”

SK Telecom

“In our collaboration with Intel, we have demonstrated how to use inference algorithms based on Intel x86 architecture to achieve better C-state power management, thereby effectively reducing network energy consumption,” said Takki Yu, vice president and head of infrastructure technology at SK Telecom.