5G

Marvell: L1 5G accelerators grab attention of telcos and vendors

  • Vendors like Dell and Nokia are using the chips, which are being deployed in a new Vodafone open RAN pilot

  • Intel had an early lead in 5G accelerators, but Marvell is now coming through

  • Marvell's senior director of product marketing told Silverlinings to expect a "barrage of announcements" at Mobile World Congress 2024.

It’s time to catch up on what is happening with Marvell’s Layer 1 (L1) 5G accelerators.

While most of the attention right now is concentrated on the development and adoption of artificial intelligence (AI) chips – and Marvell is producing AI silicon – telecom operators and vendors are starting to use 5G L1 accelerators in virtualized RAN (vRAN) and cloud RAN trials and even deployments.

So, Silverlinings talked to Joel Brand, senior director of product marketing at Marvell, about what’s happening with its 5G L1 silicon. “It seems to me that this field is only crowded in the press, because if you actually want to use something, there are only two viable options at the moment,” he claimed.

Brand asserted that vendors and operators have the choice of Intel or Marvell for 5G accelerators at the present time. He noted that Intel offered “lookaside” microchips for accelerator duties. “This is a great solution for anyone who wants to experiment with 5G, develop prototypes, and build small scale systems to test the Cloud RAN concept,” Brand stated.

Roy Chua, principal analyst at AvidThink previously said, however, that Intel has got an “early lead” in acceleration with operators using its field programmable gate arrays (FPGAs) in conjunction with Intel server CPUs and the firm’s FlexRAN software. Chua noted in May that Dish and Rakuten Mobile both support Intel’s FlexRAN architecture.

Marvell is coming up

Even if Intel has an early lead, vendors and operators are now set on deploying Marvell L1 accelerators for cloud radio access networks.

You could use any of Marvell’s customer systems, from the likes of Dell to build a system that, at least as far as the L1 is concerned, is providing the same features, performance and energy efficiency of the currently installed dedicated RAN systems, Brand claimed. “That’s the reason vendors are picking Marvell – cloud RAN is not going to happen if it’s worse than currently deployed systems,” he said.

“Nokia announced something with Vodafone last week,” Brand also noted. Indeed, Vodafone is putting together an open RAN pilot using a cluster of sites in Northern Italy. This pilot will use Nokia containerized baseband software running on Red Hat OpenShift hybrid cloud platform. All this is based on Dell PowerEdge XR8000 servers, and Dell is using the Marvell Octeon 10 L1 5G accelerators.

AvidThink's Chua told Silverlings in an email Monday that the deployment uses Dell servers with the Marvell/Nokia SmartNIC that uses Nokia's Layer 1 stack. "Marvell has been building chips for mobile communications for a long time and understands the needs of the mobile RAN, transport, and core," the analyst continued."They were already a prominent silicon supplier for traditional RAN deployments prior to the open RAN movement, and vendors like Nokia and Samsung likely picked them because of that expertise."

There's more to come on this, Marvell said. “I would expect a barrage of announcements at [Mobile World Congress in 2024],” Brand noted.

 A crowded field 

“Many are trialing the Marvel devices (and the Qualcomm devices) because they think that since they are essentially ARM based, that they will have lower power requirements than x86 for Intel (or AMD), and power is a major cost factor for carriers,” Jack Gold, principal analyst at J.Gold Associates, noted in an email to Silverlinings. “But Intel has pushed back by creating a Xeon server SoC that has a vRAN accelerator built in, so you don’t need to have a separate card, and can save power and cost of additional hardware.”

Gold said that MediaTek and probably others in China are producing 5G accelerators as well as AMD, Intel, Marvell, Qualcomm and Nvidia. It is, as he said, “a crowded field.”