Where are we in the market for silicon that boosts the performance of the virtualized radio access network (V-RAN) right now?

Enabling merchant silicon to improve layer 1 (L1) acceleration in the V-RAN is still mostly Intel’s game to lose so far, analysts told us. Still, there are questions about how the market will evolve.

There are several components in a V-RAN, Roy Chua, principal analyst at AvidThink said in an email reply to Silverlinings’ questions.

These components include the remote radio unit (RRU), which handles the physical layer transmission and reception. The virtualized central unit (vCU), the equipment that deals with non-real-time processing and access control and the virtualized distributed unit (vDU), which deals with real-time processing and lower layer protocols, such as media access control (MAC).

Chua said that most of the discussions about V-RAN acceleration center around the processing architecture in the vDU.  

“With regard to the vDU, Intel (from the original FPGA-assisted FlexRAN, to the recent 4th Gen Xeon Scalable with AVX extensions), has been the most aggressive in pursuing vRAN solutions for telcos,” the analyst wrote. “The other players have historically been less aggressive and spent less money and efforts in developing and nurturing the ecosystem.”

“AMD has shown some limited interest on the vRAN front with their EPYC CPUs, but more around Xilinx FPGAs for telco use cases,” the analyst stated. “Nvidia has also been pushing forward on both the DPU acceleration front for 5G core, and is now pushing a GPU-based approach for vRAN with their Aerial SDK.

“Marvell and Qualcomm are both in play too, using custom silicon and Arm-based cores,” Chua noted.

“However, from a general purpose CPU standpoint, I see that Intel continues to demonstrate strong interest in this market and is doubling down on this front, with the support of their customer/partners like Rakuten, Ericsson, Mavenir and others,” the analyst stated. “That will likely keep the Intel x86 architecture as a major provider in vRAN for the next few years.”

Leonard Lee, executive analyst at neXt Curve, largely agrees with Chua’s viewpoint on the silicon acceleration market for vRAN. “Marvell with their OCTEON 10 portfolio and AMD should be able to make some inroads into vRAN as some operators have expressed concerns about Intel’s dominant market share,” the analyst told Silverlinings on LinkedIn. “In particular, these new competitors will likely find luck closer to the radio, namely the DU and RU where inline accelerators will be handling the L1 processing.”

“Regardless, software is important. Intel has a leg up with FlexRAN, and a significant market share lead that will be challenging to cut in the near term,” he added.

Lee isn’t quite as certain on the prospects for Intel’s 4th gen Xeon processors yet. “It entirely depends on how Intel’s new Xeon processors perform in the wild,” he said. “There is no doubt that the 4th Gen Xeon is an interesting architectural direction that reduces the need for accelerator cards and transitions Intel away from lookaside acceleration for L1 processing.”

“It wouldn’t be a surprise to see Marvell and AMD quickly follow, especially if Intel gains meaningful traction with Ericsson and Nokia,” Lee told us. “The embedded accelerator architecture could be a thing especially if Xeon processors show up in the Cloud RAN DUs of the big guys,” he added.

El artículo original puede consultarse en: