Intel boasts open RAN monopoly as Nokia turns to others

MWC23 – Intel is certainly not shy about its monopoly status in the market for open and virtualized radio access network (RAN) technology. "Nearly all commercial vRAN deployments – 99% or so – run on Intel today," said Sachin Katti, the head of Intel's network and edge group, whose revenues grew 11% last year, to $8.9 billion, as company sales tumbled more than 20%, to $31.7 billion.

Such dominance would horrify the industry if open and virtualized RAN networks were a bigger deal. As things are, they account for a slim share of the overall RAN market. Most products still come from Ericsson, Huawei and Nokia. And those products mainly use customized silicon based on the blueprints of Arm, a company headquartered in the UK and currently owned by Japan's Softbank.

But open RAN has emerged as a telco priority in the last few years, and market research firm Dell'Oro Group thinks it could account for as much as 20% of the whole RAN market by 2027. Operators hope the new interfaces will let them combine vendors at the same mobile site – using one company's radios with another's baseband software, for instance – instead of buying all the goods from one supplier, as they generally do now.

Figure 1: A closeup of an Intel processor.
(Source: Ruslan Lytvyn/Alamy Stock Photo)

The assumption is that most of these networks will also be virtualized, which is where Intel comes in. Already commonplace in the telco core, true virtualization would end the tight coupling of hardware and software in baseband units, the servers that process mobile signals. For purists, the ideal is ensuring that RAN software can sit on common off-the-shelf equipment – featuring general-purpose processors (GPPs) instead of customized silicon. As a dominant force in GPPs, Intel could be the main beneficiary.

The opportunity for others lies in the unsuitability of these GPPs for the unique performance needs of the RAN (the clue's in the "general purpose" moniker). To address that, chip suppliers are pushing hardware "accelerators," circuitry that would more efficiently handle the baseband functions.

But the accelerators being developed by other chip companies would also threaten Intel's position in this part of the network. At stake are billions of dollars in future RAN spending. And Nokia has just given a massive endorsement to an accelerator system that Intel deplores.

Arm and dangerous

Called "inline" acceleration, the system is favored by a coterie of Arm-based chipmakers including Marvell, Nvidia and Qualcomm. It works by shifting all baseband functions onto a separate card (dubbed a network interface card or SmartNIC), taking the GPP-based central processing unit (CPU) out of the equation.

Intel originally fought back with a technique called "lookaside," which also uses a separate card while keeping the GPP in the loop. But the company is now critical of both techniques. Its latest move, announced last year, is to integrate the accelerator with the system-on-a-chip (SoC) that includes the GPP.

"Both inline and lookaside require a separate card," said Katti on a call with analysts and reporters before this year's Mobile World Congress. "With integrated, it is on the SoC and you're not having to buy a separate piece of hardware. That adds a lot of cost and power cost to the solution and by integrating you are taking that away."

But parts of the industry are highly skeptical. One telco source, requesting anonymity, said offloading workloads and using Arm-based silicon instead (for inline acceleration) would eliminate the need for a powerful Intel GPP on the baseband side. This might not only produce direct savings, but also spur competition, which is supposedly an open RAN objective. Intel's market share of 99% is probably enough to fuel interest.

Nokia is attracted to it as a virtual RAN option for several reasons. For starters, it has important advantages over lookaside, says the Finnish vendor in a new white paper, allowing capacity to be increased by adding cards independently from GPPs. Lookaside, by contrast, means reliance on GPP technology goes up as capacity is added.

Arm-based silicon is also lauded over Intel's x86 architecture for its energy efficiency, and not just by Nokia. Katti insists Intel has addressed this perceived shortcoming with its integrated accelerators. "We expect this to match or better the performance per watt of the best custom dedicated accelerator card on the market today," he said. But operators may take some convincing.

Tommi Uitto, the head of Nokia's mobile networks business group, is clearly not persuaded, even suggesting that integrated acceleration is just a rebadging of lookaside. "We have looked at the different alternatives and concluded that inline delivers better performance than lookaside even if you want to rename it to something," he told Light Reading. "It is still lookaside and has higher cost and power consumption."

Virtualization wars

The other concern for telcos is that products are not genuinely virtualized or open – as Deutsche Telekom words it in a recent white paper (PDF), that "commercially available open RAN solutions remain highly dependent on the chosen hardware." Katti has previously lashed out at inline accelerators as virtualization fakery, saying they introduce silicon dedicated to a particular task.

But FlexRAN, the vRAN reference design Intel typically provides along with its chips, is equally bad when it comes to openness, according to senior telco execs. Yago Tenorio, Vodafone's network architecture director, said it means "you have to design your software in a particular way using the instruction set FlexRAN gives you. Once coded for FlexRAN, it is not portable for another accelerator – you can only use Intel from that point on."

Uitto also rejects the criticism of inline as not truly virtualized, arguing it would free up GPPs for higher-layer processing needs (baseband is often referred to as Layer 1, with Layers 2 and 3 responsible for other RAN functions). "If you offload the traffic or computing needs from the CPUs that are best suited for Layer 2 and 3, it is more optimal," he said.

"If you load all processors with this integrated hardware acceleration then that processor is no longer a general-purpose processor. That becomes purpose built," he added. "Inline is the right balance between what is general-purpose computing and what is hardware acceleration. And this type of SmartNIC can also be used for other purposes."

Indeed, their use by hyperscalers is already widespread, according to Uitto. Nokia is now building a strategy called anyRAN around the inclusion of SmartNICs in its virtualized RAN portfolio. The pitch is that its products could work on any server (HPE has already announced itself as a commercial partner) with any virtualization platform running in any cloud environment, depending on the customer preferences.

What seems partly to influence Nokia is that inline chips sold by Marvell are identical to those it already buys from Marvell for its traditional RAN deployments. "It is the same silicon in traditional and accelerator-type deployments," said Joel Brand, Marvell's senior director of product marketing. "You get the exact same capacity and performance and set of features."

In other words, Nokia can use the same baseband technology across both vRAN and traditional RAN deployments. It does not have to shop separately for the vRAN parts and purchase them in lower volumes. "This brings feature and performance parity with the same software release management cadence," Nokia states in its white paper.

Small as the open and virtualized RAN market may be, Intel remains way out in front. It already counted Nokia as a silicon customer and remains the only commercial baseband partner of Ericsson on vRAN technology.

If the market performs as analysts predict, Intel can probably anticipate sales growth. But the industry is likely to see its first inline deployments this year. And no company can expect to hold onto a 99% market share for long.

Related posts:

— Iain Morris, International Editor, Light Reading