- Samsung says it’s taking a more pragmatic approach to AI RAN than its much larger rivals
- Its software-based architecture is ideal for supporting GPUs when an operator needs that kind of power-hungry solution
- GPUs might make sense at high-demand sites, Samsung’s Alok Shah tells Fierce
PLANO, TEXAS — Ericsson and Nokia are each pursuing different strategies when it comes to AI RAN. Nokia aligns itself very much with Nvidia while Ericsson prefers the purpose-built baseband approach.
Then there’s the Samsung way. Its approach is a little different, according to Alok Shah, VP of Networks Strategy, Business Development, Marketing and Strategic Sales at Samsung Electronics America.
“We don't always equate GPU with AI RAN. The key distinction is you don't have to have a GPU in order to get AI RAN benefits, but if there's a need for it, you want to have an architecture that's ready for it,” he told Fierce during an interview here on the Samsung campus.
If there’s a cell site where an operator desperately needs more performance – say, in the middle of New York City – maybe it makes sense to put a GPU there, even though you know it’s going to be expensive and power hungry, Shah said.
“That processing gain that you get from doing more advanced algorithms justifies the capex and opex,” he said. “For a lot of other sites, it doesn't really make sense today, so our view is we're very pragmatic about it. We want to have an architecture that lends itself to bringing in a GPU when it makes sense.”
Samsung’s proving out its virtual RAN (vRAN)/cloud RAN strategy with carriers throughout the world, with Verizon being its only big U.S. customer. Outside of the U.S., Telus, KDDI and Vodafone are among Samsung’s clients.
“All of them have moved forward with Samsung vRAN and obviously, Verizon has done it at a completely different scale,” he said, noting it’s done “tens of thousands” of vRAN deployments with Samsung.
Samsung’s history with Nvidia
Although it seemed as though everybody tried to hitch themselves to Nvidia’s superstar power during this year’s Mobile World Congress in Barcelona, it’s worth noting that Samsung has worked with Nvidia for more than 20 years, starting with supplying components like memory.
Their relationship in mobile networks doesn’t date that far back, but for sure, they’ve been partners in recent years, including when it comes to benchmarking Samsung’s vRAN software with Nvidia's products.
“They've got this CPU-GPU combination. We've done some benchmarking, both for how our software runs on the CPU and also how some of our more advanced algorithms run on the GPU, and whether we can leverage that GPU to provide a performance boost with both downlink and uplink throughput,” he said.
Samsung’s leadership in software-driven architecture, vRAN and disaggregated open RAN has been a hallmark for at least five or six years now, Shah said.
“The GPU kind of fits within that architecture very well,” he said.