This week, Vodafone announced a new partnership with Intel that will give its Malaga chip research centre access to test sample silicon from Intel. The aim is to allow smaller companies to test out integrations of new software and applications for Open RAN on Intel’s platform – something they would never be able to afford normally.
Although that not sound like much, if the programme goes ahead Intel’s commitment is likely to run into tens of millions of dollars in support of Vodafone’s desire to provide partners with chip samples with new technology integrated on the silicon.
One chip rival described to TMN the process of producing the design, the IP, producing the “tape out” of the mask for the fab to produce the samples as costing a likely $60-80 million.
Santiago Tenorio, director of Network Architecture at Vodafone, agreed that the idea is for Intel to cover the cost of chip development and integration, and producing the test samples.
“I think we will need to land the discussions with Intel in a case-by-case but that’s definitely the spirit of the agreement. Otherwise there is no news. Anybody can show up to Intel and pay for the cost of testing.”
He added that it is also in Intel’s interest to do so, as it can develop its own Open RAN platform capabilities as a result, benefitting from Vodafone’s funneling of ideas and applications to the chipset developer.
“They will attract designs to the foundry, and remember this is not mass production of chips. You are providing people with a chance of prototyping their use cases. In some cases to find investors and in some to provide verification.”
Tenorio said that the aim of the partnership is to give smaller companies the means to prototype their solutions in real silicon.
“The key thing is you are giving a way for smaller players to come with their innovation and their ideas. Some of them will come and do Software Defined Radio, some with a better scheduler idea, or with some artificial intelligence applied to Layer 1. Now the [current] path for them is really long. They need to integrate into somebody else’s accelerator, server or software.
“Imagine if they can come to Malaga to work with us: we can funnel ideas, select algorithms or IP that looks promising, then we can send it to Texas and it comes back with integrated silicon – x86 or a powerful FPGA or anything else, it doesn’t have to be a full CPU. So it comes with enough sampling that you can use for prototyping. Then you build something that looks like the final product and assemble the final use cases and test it, take it to the field, benchmark and follow up and see if the promise is delivered or not. What you just created is a path of zero resistance with no relevant cost to a true prototype.”
Under ARM server
At FYUZ, Vodafone also announced that it would be experimenting with ARM on the servers that run the OpenRAN software and cloud stack.
To date most of the discussions about processor performance for vRAN has focussed on the hardware acceleration for lower layer radio processing – so called Layer 1. This has been where the “never-ending” (in Tenorio’s words) arguments about inline vs lookaside acceleration has resided, and where earlier FPGA-based approaches from Intel and Xilinx have been challenged by ARM-based players such as Marvell (which powers Nokia’s L1 DU SmartNIC), Qualcomm and nVidia GPUs, as well as by Intel’s more recent vRAN Boost design on its Sapphire Rapid platform.
But Vodafone’s recent announcement is not just about this L1 element, rather it is about the design for the entire server that might host the Du, or DU-CU.
Vodafone’s Santiago Tenorio said, “As much as we love working with Intel, our strategy is not to have a single supplier in anything we do. As of today if you look at Open RAN, in reality the servers and CPUs are Intel. Some of the most prominent accelerators are based on ARM and that’s a good thing. But today’s announcement is about the lack of a practical, commercially available, ARM based server.”
Tenorio said ARM is not far from a commercial server but Vodafone wants to do more in terms of performing lab and field trials, stacking the server with different OS, containers and app software.
“Then we will benchmark it for performance and cost. There are lots of expectations and statements out there on the performance and cost of ARM based servers, compared to an [Intel] Ice Lake or Sapphire Rapids. OK , it’s time to see it in action.”
So while it’s true that a new server brings optionality it also brings new SI challenges. That’s why Vodafone also announced that Fujitsu will be the first to write code for the ARM CPU server.
Dual stream required
Talk of optionality raises the issue of “lock in” to the Intel FlexRAN platform, whereby the level of hardware-software integration means that once a vendor and operator is on the FlexRAN path, it is difficult to port away. This is something that Rakuten’s ex-CTO raised in the past, for example, even as he committed to building Rakuten’s network and Rakuten Symphony products on the Intel platform.
Tenorio: “I think in general people have hesitated to put too many eggs in a single basket. And they may have been scared of fully adopting and embracing the FlexRAN instructions, ending up locked in to an ecosystem. But I think the industry is starting to realise that you may have to keep one software stream going for a few years to come, instead of betting on a single horse. So we’ll see where it ends up.”
One possible longer term landing may be further abstracting software from hardware.
O-RAN Alliance’s Working Group 6 in O-RAN Alliance is working on an abstraction layer that would, in theory, end the software/hardware dependency for L1 software.
“We’re maybe a year away from seeing the first efforts on abstracting L1 come to fruition. I think that will help” Tenorio says. “Until that matures, then we may want to have a couple of streams on two different hardware to see how things evolve.”
