It turns out that we are wrong all the time.
We were all told that AI in itself, 5G in itself, and computational processes in themselves had to cause monumental changes in computing. As a result, we have built expectations around what each of these technologies should have given on its own, but honestly, the real results are disappointing.
People are beginning to realize that you need all three to work together at the same time to feel their full impact. It’s a combination of the speed and low latency of 5G, plus the extended range of computing computers and AI intelligence that can power the types of impact applications and futuristic scenarios we all originally promised with 5G.
To her credit, Nvidia seems to have recognized this moment some time ago. The company is using this year’s MWC to explain some of the critical details of its strategy in these areas, using the previously announced efforts of the Aerial A100. In particular, the company described its AI on the 5G platform, highlighting both the software development and the semiconductor roadmap needed to revitalize AI-driven, delivered 5G end-to-end computer applications, while enabling vRAN and Open RAN functionality – all this within a single server.
There’s a lot to unpack here, so let’s start with the basics. Nvidia has been offering enterprise-focused, GPU-accelerated AI software for some time through its EGX platform and its CUDA-based software development tools. A few months ago, the company also announced Aerial 5G, a software-based development platform designed to allow the Open RAN software stack, virtualized network functions (VNF) and other telecommunications-related workloads to run faster on Nvidia GPU-equipped servers.
The Aerial A100 is an accelerator plug-in for x86-based servers, which includes an Ampere generation GPU and a BlueField DPU (version 2.0). In combination with Aerial 5G software, the combination allows you to create a software-defined part of a 5G network infrastructure using COTS, or commercially ready servers from companies such as Dell, HPE, etc. Because the card is identical to another used to speed up servers focused on AI applications running on Nvidia’s EGX platform, the exact same server can work with CUDA-powered artificial intelligence applications.
As it turns out, the parallel processing capabilities of GPUs, which makes them so suitable for running applications based on AI and machine learning, are also very suitable for running a fully compatible with O-RAN 7.2 5G RAN software stack capable of full acceleration of layer 1. In fact, together with the BlueField-2 DPU, which is optimized to efficiently move parts of data into cloud-style workloads, end-of-life computer styles, the combination of antenna card and software creates fast, powerful network infrastructure equipment. This combination supports what Nvidia calls 5T for 5G (Telco time-sharing technology), which is the ability to support the time-sensitive version of SyncE on Ethernet, allowing low-latency computer applications. Using software RAN stacks from partners, including Mavenir and Altran, Nvidia’s platform can also support full Massive MIMO 64T / 64R-compatible radio and antenna systems via a standard eCPRI connection.
On a more practical level, the ability to process RAN and AI software applications in one box also makes it easier to deploy a server in multiple environments. This in turn makes it suitable for widely dispersed end computer applications. Also, for businesses that may want to deploy private 5G networks, combining the two options can potentially reduce costs and energy consumption, as well as overhead network management costs.
Of course, the world of software-defined, O-RAN-based network infrastructure is relatively new, so there is a lot of training and experimentation. As a result, it was interesting to see that Nvidia also announced a partnership with Google Cloud to create a joint lab focused on creating applications based on AI-on-5G. As the telecommunications world begins to move away from the traditional closed, customized infrastructure boxes that have powered it over the past few decades, it embraces modern container-based software architectures, pioneers of major cloud computing providers. Last but not least, it’s surprising to see companies like Google Cloud (as well as Amazon’s AWS and Microsoft’s Azure) engage more actively with traditional technology companies like Nvidia to work together to create important new offerings for the network infrastructure market. . . In fact, I expect this type of partnership to expand rapidly and to expand significantly over time. These combinations of companies are likely to stimulate the creation of advanced applications that use both 5G, AI and end-to-end computing.
The latest piece of Nvidia AI-on-5G news focuses on its silicon roadmap for the telecommunications market. The next version of the Aerial series cards is expected to include both the next generation GPU architecture chip and the BlueField-3 DPU. Together, Nvidia said the combination should offer three times the wattage performance of vRAN-based loads than today’s model. More importantly, the BlueField-3 will include sixteen A78-based Arm CPU cores. This means that the card can run on its own on a server without an x86 processor – in other words, it allows for an Arm 5G-based network infrastructure. Nvidia explained that it will also function as the current card as a plug-in for x86-based servers, but the standalone option is certainly an interesting development – especially in light of Nvidia’s desire to buy Arm. The company then plans to integrate GPUs and DPUs on a single chip with further improvements in watt performance.
Ultimately, what Nvidia demonstrates by publishing these additional details about its Aerial 5G offerings is that the combination of chip technology as well as software tools position it as a potentially great player in the world of 5G network infrastructure. It’s still relatively early for O-RAN-based software environments, but it’s clear that Nvidia is making a serious and compelling effort to expand the company’s influence beyond its current position.
Disclosure: TECHnalysis Research is a market research and consulting firm in the technology industry and, like all companies in the field, works with many technology providers as customers, some of which can be listed in this article.