Summary Bullets:
• Nokia will use CPUs from Nvidia to bring more choices to the vRAN space, following the former’s ‘anyRAN’ concept.
• Nokia also aims to use Nvidia’s GPUs to transform telecom networks using AI – an aspirational future vision.
Nokia announced this week that it will collaborate with chipmaker Nvidia in two ways – both focused on virtual radio access networks (vRAN). Nokia will use Nvidia Grace CPU Superchip for processing vRAN functions at Layer 2 and above, together with Nokia’s RAN software and hardware accelerators (PCIe cards that boost vRAN performance). It will also use the graphics processing units (GPUs) Nvidia is most known for to handle AI applications and vRAN acceleration.
The use of Nvidia Grace CPU Superchip stems from Nokia’s (and the broader vRAN sector’s) interest in expanding and diversifying the vRAN silicon ecosystem, which has been largely dominated by Intel thus far. That’s especially problematic for Nokia because Intel favors accelerators, with what’s called ‘lookaside’ architectures, that Nokia does not prefer. Nvidia’s CPU is based on an Arm design – an alternative to Intel. Nokia has also been working with other silicon suppliers, including Qualcomm and Marvell, but the more players in this space (especially heavy hitters like Nvidia, a roughly trillion-dollar company), the healthier the ecosystem will be. Diversification also feeds Nokia’s ‘anyRAN’ concept, wherein it uses partners like Dell and RedHat to give operators more choices for RAN and vRAN infrastructure.
The use of Nvidia’s GPUs has a separate trajectory. RAN vendors have been interested for years in the potential to use GPUs for vRAN. In fact, Nokia’s rival Ericsson announced a partnership with Nvidia to do exactly that in 2019 (though Ericsson never commercialized GPU-based vRAN). In particular, GPUs are interesting for two reasons: (a) their ability to handle multiple workloads simultaneously and (b) their ability to handle AI applications. That multi-tasking capability could fit vRAN well to the extent that solutions are developed to bundle vRAN with other applications – such as edge computing, packet core, etc. – especially to serve enterprise needs. But it’s always been much less clear how well-suited GPUs would be for vRAN alone, especially in terms of power efficiency.
Nokia is leaning hard into the AI aspects, claiming that GPU-based vRAN will lead to something it calls ‘AI-RAN,’ a fundamental transformation of telecom networks that the vendor didn’t concretely define.
The GPU side of this deal appears to be more aspirational, while the CPU side is more practical. Nokia hopes to validate a vRAN proof of concept based on Nvidia’s Grace CPU in 2024 and work toward productizing in 2025 and beyond. As for GPU vRAN and ‘AI RAN,’ the vendor didn’t offer specific timetables. And recalling Ericsson’s 2019 Nvidia announcement, it would likely take quite some time for the vision that Nokia hailed this week to become reality.