As the industry heads to MWC next week, Intel has laid out a strong defence of its position within telco vRAN deployments, as it “introduced” its Xeon 6 SoC, formerly known as Granite Rapids-D.
Making sense of Intel’s multiple platform “launches” can be tricky. As well as today’s launch at which the Xeon 6 SoC was “introduced”, Intel had previously announced the Xeon 6 SoC back in November 2024, publishing a lengthy blog post on the new platform’s capabilities. Today, Intel is launching the 6500 and 6700 series of its XEON 6 platform, the processors on which the XEON 6 SoC platform is based.
But timelines aside, the messaging is clear. Because it is the most integrated platform to date, XEON 6 SoC reduces TCO for telco AI-optimised network function workloads. It enables single server deployment of workloads such as vRAN, integrates accelerators such as vRAN boost to enable CPU-based acceleration, and bundles AI technologies to support AI functionality within the network without any additional discreet accelerators.
XEON 6 SoC is Intel’s upgrade on the XEON 4 SoC which first included its vRAN boost as an integrated, CPU-based acceleration for vRAN workloads. That platform came to market in early 2024 and has “deployment commitments from AT&T, Telus, Verizon, Vodafone and other tier-1 operators,” according to the company blog post.
There’s little doubt that Intel dominated most early vRAN deployments, with several new Open RAN entrants and also Ericsson basing its own products on Intel silicon. But others, such as Nokia, have chosen to use Arm-based inline acceleration platforms from Marvell. Some operators and vendors too are making noises about the need for true L1 software stack abstraction, so that RAN software is not in any way integrated with or dependent upon the underlying platform. With Nvidia also being increasingly noisy about the potential for GPU-based cloud and vRAN, and ramping up the AI-RAN vision at MWC 2024, there is plenty to occupy Intel’s mind in the cloud and vRAN space, never mind the maccro issues surrounding the company.
Xeon 6 SoC enables AI workloads on the CPU without the additional cost, heat and space of external AI accelerators. Ultimately CPU-native inferencing provides better TCO for most enterprise, computer vision and AI/ML workload
Presenting the new platform to media and analysts, Cristina Rodriguez, Vice President Network & Edge Group, GM Wireless Access Network Division, was clear that the company has the right solution for vRAN deployments. She said Intel was “perfecting a CPU lineup built for networks and edge in the era of AI automation.”
Rodriguez said “Operators have been very clear. They need meaningful improvement on TCO. They need more capacity and they need an affordable, open path forward to AI.”
The new SoC has up to 72 Performance Cores (CPU core SKUs counting above 42 cores are available from Q4 2025), which is more than twice the core count of the previous generation, along with improved I/O and PCIe 5.0 support to increase vRAN capacity, and integrated Intel Ethernet. It also includes integrated AI acceleration with Intel Advanced Vector extension (AVX) and Advanced Metric Extensions (AMX). And it still includes Intel’s vRAN Boost accelerator. All of that, Intel says, adds up to a 3.2 RAN AI performance per core, a 2.4x increase in vRAN capacity and a 70% improvement in energy performance.
Rodriguez said, “With these, Xeon 6 SoC enables AI workloads on the CPU without the additional cost, heat and space of external AI accelerators. Ultimately CPU-native inferencing provides better TCO for most enterprise, computer vision and AI/ML workloads. A combination of built-in AI acceleration and Intel-optimised AI tools and frameworks, such as vRAN AI Development Kit are enabling simpler introduction of AI for operators using the same hardware running their other workloads.”
Intel lined up a list of supportive customers and partners for the new SoC, including Samsung, Ericsson, HPE and operators AT&T, Verizon and Vodafone, who are all big Ericsson and/or Samsung customers. Rob Soni, Vice President, RAN Technology, AT&T, and a huge Ericsson customer provided the following quote, “We’re excited to begin deploying this year using Intel’s Xeon 6 SoC, a high-capacity, programmable virtual RAN hardware platform. This technology allows for ongoing AI advancements through seamless software updates.”
Rodriguez said Intel had received “tremendous excitement” from operators about Intel Xeon 6 SoC – the results of which would be apparent during next week’s MWC.
AI in the RAN
It is worth remembering that when Intel talks about enabling AI-optimised telco workloads, it is talking about being able to run AI in silicon to optimise the workloads themselves. It is not proposing the “rented out compute capacity” model that the AI-RAN Alliance and its big backer, the GPU-toting Nvidia, is putting forward.
But while Rodriguez says that the XEON 6 SoC can enable CPU-based, AI-optimised vRAN on a single server without the need for discreet hardware acceleration, it’s only natural to wonder what her view is on the work going on with others within the AI-RAN Alliances, which includes Nvidia and Qualcomm as members, but not Intel.
Asked the question, she played a pretty dead bat, reiterating instead the benefits, as she sees it, of the new Intel SoC.
“I’m very excited about the future of the RAN, especially with the integration of AI, which will undoubtedly bring new opportunities. While I won’t comment on the specific business case mentioned, as there are many models out there, I want to emphasise we are committed to delivering the best TCO and performance per Watt value to our customers.”
XEON E-Cores for the 5G core
At the same time as introducing the XEON 6 SoC, the company also highlighted the benefits of its new XEON 6 e-core processors, the 6500 and 6700 series, which it says have been achieving good performance and results for service providers deploying cloud core functions on servers based on the Intel hardware. These include Ericsson’s CNIS portfolio, the Dell PowerEdge R670 and the HPE ProLiant Compute Gen 12. Nokia has also collaborated with Intel to integrate E-core functionalities on its 5G Packet Core Network Function. Samsung’s next-generation release of Samsung Cloud Native Core, planned for the second quarter of 2025, will feature Intel Xeon 6 processors with E-cores. SK Telecom will deploy Hewlett Packard Enterprise DL340 servers based on Intel Xeon 6 processors with E-cores for its 5G core deployments in 2025.
