Nvidia set out its 6G stall
Although much of the platform can be used for testing 5G
Analysts said that Nvidia is looking to get its GPUs deeply into the wireless picture
Artificial intelligence (AI) darling Nvidia laid out some of future radio access network (RAN) plans this week, unveiling a 6G Research Cloud platform at the GTC conference.
Nvidia's aim partly appears to center around making its high-performance graphics processing units (GPUs) silicon as crucial to 6G RAN — and even 5G wireless — as the chips have become for many AI systems. Nvidia hasn't yet become as big a player in RAN systems as it would evidently like to be.
The platform consists of a digital twin to enable users to simulate accurate radio environments for 5G and eventually 6G systems; a software-defined, full-RAN stack that allows researchers to customize, program and test 6G networks in real time; and a neural radio framework that uses Nvidia GPUs to train AI and machine learning models at scale.
"Really there's nothing that limits its use in 5G," said senior vice president of Telecom at Nvidia, Ronnie Vasishta. Noting that 5G spectrum is being used presently.
Vasishta said the digital twin will create "a much faster feedback loop" than currently available when testing specific environments that are physically accurate. "It works from one tower to city scale," he said, noting that the twin can create an RF environment which models grass, concrete and buildings.
"You can now test your algorithms directly into that simulated environment," he said. "Over time, you'll be able to do this in real time."
No word yet on what a simulated RF environment in NYC would look like or if they can model pizza rat!
Vasishta said that the research platform is already available. "It's in limited availability right now," he said. The Aerial digital twin will be available in April.
Analysts chip in
Analysts had a lot to say about the cloud platform, as it combined two of the hottest buzz words in the industry: AI and 6G.
"A lot of its work to date has been to sketch out potential 6G architectures and what this might mean for how you implement a future RAN system," Gabriel Brown, senior principal analyst, mobile networks at Omdia, said of the Nvidia platform. "There is a play in 5G and 5G Advanced, but more as a testing ground."
Roy Chua, principal at AvidThink, told Fierce, "The current work that they describe is based around O-RAN architecture, and...it could be used to improve today's 5G RAN – so you can leverage it to explore improving xApps and rApps, and improve the performance of RAN today," he said. "They are trying to help entrench GPUs into the 6G ecosystem, particularly around the RAN, and getting it going as university and vendor research teams are exploring ways to improve massive MIMO, L1 and L2 improvements for 6G."
Nvidia clearly wants to make GPU chips the silicon of choice for 6G RAN, in a way they haven't been for 5G systems.
"They've had limited traction in the RAN vendor community so far, but with the increased interest in O-RAN, plus spillover hype in AI/ML from GenAI, they likely see this as an opportune time to push into the telco market – they look like they have support from the RAN vendors this time," Chua said.
"Taking the 6G research angle allows them to drive use of GPUs in 6G research as algorithms are being designed and then hopefully become entrenched into production platforms later," he said.
"As we always say, the RAN pie is fixed over time but it is not fixed in opportunities," Stefan Pongratz, VP at Dell'Oro Group said of Nvidia's 6G play. "In contrast to other hyped technologies, there is a growing realization that AI, vRAN and automation are here to stay," he concluded.