Nvidia is lining up with other chip designers, such as Intel and Marvell, to compete in the growing layer 1 (L1) acceleration market for virtualized radio access networks (vRAN) using a software-based GPU system that can undertake other acceleration tasks, according to the company.
“We have built a full software-defined L1 acceleration,” Soma Velayutham, global business development lead for telco at Nvidia told Silverlinings. All of the L1 acceleration is done in software.“That's the magic,” Velayutham said. “So what we're able to achieve is what Ericsson does in an ASIC architecture. Or what Marvell and Qualcomm achieve in a SOC architecture.”
“We’ve been able to take that and really deliver it in a general purpose accelerator platform,” Velayutham continued. “What that means is, if you don’t like that server you’re using for 5G, you can reboot it and go and play games with it.”
Nvidia is the only company who has come out and offered a multi-purpose system, Velayutham claimed. “Why is this so important?” He asked. Telcos worldwide spend “$2 to $3 billion dollars” on RAN compute a year. “The average utilization of the RAN compute is 30%, " Velayutham said. “You have to design for the peak.”
“This is one of the worst business cases for telcos because they have to invest in a 747, but they only have the passengers for business class,” Velayutham said. But by law you have to design the economy seats.” he added.
“Whenever you’re not using the economy seats, you can actually reform it and use it for something else," he said.
Nvidia noted in their recent blog on their new AI-on-5G system that the acceleration can be used for other tasks on the same server, such as immersive graphics and digital twin workloads.
Velayutham said that its 5G acceleration is being used by Fujitsu, Radisys, Capgemini and Mavenir in current or forthcoming products.