Featured Story
Verizon revamps FWA with mmWave 'breakthrough' to match fiber speeds #
Verizon reveals 1-Gig FWA service using millimeter wave to add 10 to 15 million high-speed broadband premises as overall Q3 subscriber growth surges.
Virgin Media O2 has installed an Nvidia processor at a busy Nokia site in central London to see if it offers genuine AI benefits.
Inside the M25 motorway, the hellish road encircling London, parts for the radio access network (RAN) operated by Virgin Media O2 (VMO2) come entirely from Nokia. They include the custom chips that handle conventional RAN algorithms, co-developed by Nokia and silicon partner Marvell Technology. But at a busy site in Paddington, world famous as the entry point into the UK for a fictional Peruvian bear, the telco has been investigating the benefits of a graphics processing unit (GPU) made by Nvidia.
That follows a new partnership Nokia struck with the giant chipmaker about a year ago. Under the terms of that deal, Nvidia has invested $1 billion in Nokia, while Nokia has been redesigning its 5G products to work on GPUs instead of Marvell's silicon. Since the advent of generative AI (GenAI), those GPUs have become the default choice for developers training large language models (LLMs) in data centers. In the RAN, Nokia claims they are allowing it to boost spectral efficiency and make other valuable improvements by replacing traditional algorithms with AI. It needs telcos to agree.
What VMO2 discovers in Paddington, therefore, is critical. Numerous other telcos are conducting similar GPU RAN trials this quarter before products are due to launch commercially next year. Hardly any have so far had much to say about the experience. But massive gains in performance or efficiency could spur companies to invest in GPU-based products as an alternative to other forms of network upgrade.
The VMO2 trial has specifically focused on the radio conditions that exist between the antenna at the mast and the smartphones used by customers in the Paddington area. "What we've been working with them on is using AI to better estimate the channel quality," Rob Joyce, VMO2's director of mobile access engineering, told reporters and analysts at a recent briefing.
Traditional algorithms work according to predefined sets of parameters about radio conditions. AI essentially learns on the job, like a precocious apprentice, and adapts quickly to specific real-time factors. "You optimize according to what you see, and it trains itself on the environment the cell site is in," explained Joyce. "So, the basestation tunes its algorithms to that environment."
The result seen by VMO2 in Paddington was a 20% improvement in spectral efficiency, meaning the site was able to carry more data without needing additional capacity or spectrum. Joyce's assessment seems to validate some of Nokia's own claims about GPUs, and VMO2 expects to see even more substantial gains in the future.
But a massive rollout of GPUs across the London area and other parts of the Nokia footprint would be a surprise at this stage. For one thing, VMO2 has already spent about £1.4 billion (US$1.9 billion) on mobile network upgrades in the last two years, installing new radios from both Nokia, its partner in the southern part of the country, and Ericsson, which it uses in the north. The £700 million ($940 million) tranche of spending announced early this year does not include any plans for a GPU RAN deployment, Joyce confirmed to Light Reading.
More generally, many telcos and industry experts harbor concerns about putting GPUs into the RAN. Hardly anyone believes that chips promoted by Nvidia as a "general purpose" technology can ever be as power efficient as custom silicon designed, by definition, for one purpose. The GPU RAN equation needs to show that other AI-related benefits can outweigh those costs.
GPUs, however, are no prerequisite for using AI in the RAN, according to Joyce. "You don't need a GPU to do this," he said, when presenting the results of the Nokia trial. "I think, to be fair, that Ericsson silicon is very good as well."
In contrast to Nokia, the Swedish vendor continues to market its own application-specific integrated circuits (ASICs) as a suitable RAN platform for AI. And it already claims to have AI-capable ASICs installed in commercial networks. Around 60,000 basestations operated by its customers now support an AI successor to link adaptation, a RAN algorithm, said Per Narvinger, Ericsson's CEO, on a recent earnings call. It is boosting spectral efficiency by around 10%, he said.
High-capacity Ericsson ASICs such as the RAN Processor 6672 are naturally designed for inclusion in Ericsson's own appliances, the pizza boxes traditionally installed at the bottom of towers. They can also be pooled in a centralized RAN facility that supports multiple radios. But common, off-the-shelf servers may offer more scope for this kind of aggregation. Dell, Quanta and Supermicro are all now incorporating RAN-capable GPUs into servers that could be used for what Joyce describes as a "cloud RAN" deployment.
"If you've got high-capacity fiber, you can move that GPU to a centralized location" to control a number of sites, he said. "That's beneficial because you could then do what we call coordination and scheduling between multiple technologies and improve that even further."
His concern is the fiber. The Paddington site is connected to 10Gbit/s fiber, providing more than enough capacity for local needs. Centralizing the baseband would require VMO2 to connect that facility back to its radio sites with a technology called CPRI (common public radio interface), or its eCPRI (evolved) successor. "Each one of those is 25 Gbit/s," said Joyce. "You may end up then having to run three CPRI connections, so 75 Gbit/s to the tower, as opposed to the 10 Gbit/s."
Almost inevitably, that would mean adding fiber or paying more for transmission, explaining Joyce's reservations about investing in a centralized RAN. "My view is still that I think the distributed RAN is the way forward, and, in that case, if you're going to get this, you're going to have to put additional baseband GPU resources at the site."
For Nokia, there is a risk customers will continue to resist an upgrade that would demand the rollout of new hardware, whatever it may cost. Nokia still offers products based on Marvell's silicon, and will do so for years, but it is no longer developing new technology with Marvell. If Nokia is to have a long future as a RAN vendor, customers will ultimately have to adopt GPUs. There would be danger for Ericsson, perhaps, if its current ASICs could not accommodate emerging AI features, which today's powerful GPUs may already be able to support. Customers now benefiting from link adaptation improvements on hardware they have used for years might discover it needs to be replaced sooner than expected for more advanced AI.
Telco anxiety about either vendor's technology roadmap could be an opportunity for the other to grow market share. But with so few RAN vendors in the market, operators desperately need both Nordic companies to succeed.
Read more about: