AI

Vodafone Tests AI-RAN Efficiency Amid GPU Debate

LinkedIn Google+ Pinterest Tumblr

Vodafone is testing whether AI can make mobile radio networks more efficient. The work takes place at its innovation campus in Malaga, Spain. The city offers a useful backdrop. Summer crowds often create heavy mobile traffic. That mirrors pressure on busy urban networks.

At the center is AI-RAN, where software learns to improve radio decisions. Traditional RAN systems use fixed algorithms for tasks like signal management. AI-RAN replaces some routines with adaptive models. Operators hope this can raise spectral efficiency. That means sending more data through the same airwaves.

Kyriakos Exadaktylos, Vodafone’s head of RAN evolution, sees strong potential. “I think the route to improve is quite significant, and AI will help for that,” he said at an Omdia AI-RAN symposium.

However, the hardware question remains open. Nvidia has fueled much excitement around AI-RAN. Its GPUs dominate many AI data centers. So, some vendors view them as natural AI-RAN engines. Yet Vodafone remains cautious. GPUs can cost more and draw high power.

“We need to go into our labs that we’re now building with advanced capabilities to test the benefit on network efficiency,” said Exadaktylos. “So, we have not, at the moment, stamped that we have a green light to go to deployments with GPUs.”

The alternative is to run AI on existing silicon. That includes CPUs, Arm-based chips, and custom ASICs. ASICs are purpose-built chips for specific workloads. They can execute selected radio tasks quickly and predictably. This matters because RAN decisions happen in microseconds.

Ericsson argues that custom silicon still has a strong role. “The really critical part here is that you have about 40 microseconds before you need to make a decision on link optimization,” said Gabriel Foglander, Ericsson’s head of RAN strategy.

Still, custom chips face a business challenge. The radio equipment market is smaller than data-center computing. That limits returns for chip designers. If AI-RAN changes quickly, ASIC refresh cycles may become painful. A stronger GPU could last longer in commercial networks.

Vodafone also sees room in software. “We think that at least those that are existing algorithms, and will have AI embedded, will not need a change on silicon,” said Exadaktylos. More advanced algorithms may eventually change that view. “Maybe it’s two, three years down the road, and we will need to validate.”

Nokia is also under close watch. Nvidia invested $1 billion in the company last year. Nokia plans future RAN products around Nvidia GPUs. Yet a recent Nokia and Marvell demo used existing Marvell silicon. It reportedly improved spectral efficiency by 20%.

The debate also affects 5G and 6G planning. Vodafone believes meaningful gains may remain possible with OFDM. This waveform sits at the heart of 5G. It should also continue into 6G, reducing forced hardware overhauls.

Malaga may therefore influence real network choices. Vodafone relies mainly on Huawei and Ericsson across Europe. In the UK, VodafoneThree selected Ericsson and Nokia for thousands of sites. The lab results could shape future AI-RAN deployment across major markets.

Write A Comment