NVIDIA wants to reshape the telecom industry with its GPUs

NVIDIA is one of the companies that has best managed to diversify in today’s tech industry. The brand traditionally known for GPUs established itself as a pioneer in the segment of artificial intelligence chips. Now, NVIDIA wants to achieve something similar in the telecommunications segment, having ambitions to reshape the 5G industry with its hardware.

The telecommunications segment operates almost independently from the rest of the tech industry. Developments based on commonly used architectures are not usually directly transferable to mobile networks. This is because the latter are built around technology focused on a single, very specific purpose. In other words, hardware dedicated to networks is limited only to that area, without any capacity for adaptability.

NVIDIA’s ambition is for its GPUs to be the basis of the 5G of the future

NVIDIA identified the issue and is positioning itself as a potential driver of future mobile network architectures. The company has already successfully adapted its GPU business to AI chips, making it the current world leader. Now, NVIDIA’s ambitions are for its GPUs to replace telecom’s 5G network hardware platforms.

NVIDIA’s vision is in line with that of IT companies such as Intel and others. They have been promoting the creation of a “general purpose” hardware platform for some time, which can also serve as the architecture for mobile networks. The goal is to replace current equipment with hardware widely present in the tech industry. For example, with the same chips that are present in the computing and server fields.

The high adaptability of a potential new platform for telecommunications networks would transfer the economy of scale of products from the IT sector. This would translate into better prices since telecoms would not have to resort to hardware developed specifically for a niche much smaller than the general IT sector.

In addition to the economy of scale, the adaptability of the new platform could be another big advantage. Telecoms could double the utility of their mobile network architecture by enabling the hardware to run generic workloads at the same time.

Cloud-based RAN products with CUDA architecture

With that in mind, NVIDIA wants to adapt its GPUs as RAN (radio access network) chips. However, the company’s focus is offering cloud-based RAN products with its CUDA architecture. NVIDIA wants to fill the vacant slot of general-purpose hardware that is also suitable for telecom platforms.

The brand claims that its cloud RAN chips based on CUDA can offer similar or better performance than hardware developed for a specific purpose (ASIC). “What you’re really getting is the performance of, or even better performance than, an ASIC that is actually in a software-defined architecture,” said Soma Velayutham, the general manager of NVIDIA’s AI and telecom business. “CUDA is almost like a software-defined ASIC,” he added.

Some obstacles to overcome along the way

NVIDIA still has hurdles to overcome before it can achieve its ambitions in the 5G segment. First, the industry today seems convinced that GPU-based technologies are quite power-hungry. AI-based developments have been shown to require large amounts of power, leading big companies to build dedicated power supplies. They believe this factor will result in more expensive RAN products.

NVIDIA argues that its RAN products can perform multiple complex tasks at once. Not only could they form the basis of carriers’ mobile network architecture, but they could also function as AI accelerators. Plus, the company says telecoms can choose from multiple GPU options according to their needs. “We have a range of GPUs, it’s a sliding scale,” Velayutham said. This could help balance—or even reduce—costs compared to current ASICs.

There are also certain cases where distance prevents the implementation of NVIDIA’s RAN products. Sharad Sriwastawa, CEO of Rakuten Mobile, told reporters that the technology is not economically viable when there are 20 kilometers between the DU (distributed unit) and the radio site. NVIDIA responded to this by pointing out that telecoms can deploy low-power GPUs at the radio site in remote rural areas. There is a wide catalog of hardware that could be adapted to the particular needs of each case.

Currently, Intel dominates the RAN segment, so telecoms may prefer to turn to the name with more proven reliability. This will surely complicate NVIDIA’s plans regarding its CUDA-based cloud RAN products. However, a potential aggressive pricing strategy and the added benefits of general-purpose rather than specific hardware could help convince some big names.