Your browser is not supported. Please update it.

17 April 2020

Ericsson seems to gain confidence in its GPU-based vRAN development

As the major RAN vendors develop virtualized platforms, they have been forced to engage in a significant reassessment of their base station chip strategies. The economics of the cloud-based network give them the opportunity to save cost by adopting general purpose processors (GPP) from Intel or an ARM partner; though general purpose architectures also rob them of the differentiation they had while still designing many of their chips inhouse or through captive semiconductor vendors. But as Nokia’s U-turn on its 5G base station silicon showed late last year, the cost advantages of merchant, partially off-the-shelf solutions are hard to ignore.

But some RAN processes remain beyond the capabilities of current GPP and must be supported with specialist, even proprietary chips; and/or accelerators based on FPGAs or GPUs, to work alongside the central processors. Last October, Ericsson surprised the industry when it announced a project with Nvidia, to explore the potential to use GPUs (graphical processing units) for the more demanding base station tasks, especially those taking place in Layer 1 (the physical layer) of a virtualized RAN, and in distributed base stations with low latency requirements.

At the time, despite predictable enthusiasm from Nvidia – the market leader in GPUs – for this new potential (and high value) market, Ericsson was careful not to commit itself. It found the idea of a GPU-based Layer 1 sufficiently interesting to devote time and funding to the project, but remained sceptical that anything but dedicated, single-purpose silicon could do the job optimally.

That was the overall narrative, as summarized by Thomas Noren, Ericsson’s head of 5G commercialization. He said Ericsson’s existing baseband, running on specialized silicon, could already handle the “incredibly computer intensive” software needed to support a demanding 5G workload such as dynamic spectrum sharing (DSS). In terms of replacing that with the Nvidia platform, he was merely “open to the idea”. He said: “We will explore if we can develop a distributed baseband product (DU) with Nvidia GPU. We don’t have any committed product plans, but we think this is a very interesting idea.”

If that was hardly a ringing endorsement, a new blog by Matthias Riback, head of new architecture for Ericsson’s Product Area Networks, is more upbeat (https://www.ericsson.com/en/blog/2020/4/hardware-acceleration-5g-vran?).

Riback wrote: “We see clear advantages in implementing the baseband processing and new emerging 5G use cases in close proximity, with both benefiting from Nvidia GPU-based hardware acceleration.”

He was referring to a virtualized, distributed RAN in which GPP-based hardware would enable “a wide range of emerging edge applications including augmented reality, AI and deep learning”, while scaling down to meet the needs of private networks. For the Layer 1 processes, the GPUs would provide necessary acceleration.

By contrast, for the physical layer of the vRAN, gains from just using multicore CPUs would not scale linearly, “meaning the first few cores beyond the initial one may cost more in lost per-core performance than is gained in parallelization. On the other hand, with Nvidia GPUs you can take an extreme multi-threaded approach from the start,” Riback explained.

And the focus on the edge reflects the way that, in Ericsson’s view, operators could steal a march in edge computing. They need to build out distributed hardware to support vRAN anyway, so by including edge capabilities and services, they can achieve a much enhanced ROI, compared to a company which is deploying only the edge infrastructure. In turn, they can provide infrastructure on which developers can test out their edge applications, building confidence and interest in the nascent market.

The publication of the blog at this stage suggests that Ericsson is feeling sanguine about its tests so far, compared to its cautious mood six months ago, and perhaps sees a chance to seize further high ground in terms of base station chip architectures, over Nokia, which is still smarting from its U-turn late last year.

Nvidia’s attack on the vRAN is the latest in a long-running battle with Intel’s CPU-centric approach to high performance processing, which was one of the defining battles of the twentieth century chip industry. A GPU-first strategy has scored highly for Nvidia in AI, XR and supercomputing, but before Ericsson’s project, was scarcely visible in the vRAN.

But not only did Ericsson announce its work with Nvidia in October, it also talked about a cooperation with Verizon to support integration of edge computing with a distributed, disaggregated vRAN, with a heavily GPU-led approach.

This is all about finding chip technologies which really can cope with the extremely high demands of vRAN workloads and deliver performance that is equal to, or better than, a dedicated architecture. Doubts that this will be possible using off-the-shelf chips are one reason why operators have been so slow to embrace vRAN, at least outside indoor and small cell networks.

Nvidia claims it is applying recent breakthroughs in GPU technology for supercomputing to the vRAN challenge and has worked with Ericsson to build the “world’s first software-defined 5G RAN”. Its new EGX Edge Supercomputing Platform is designed to be flexible, to support a variety of high performance use cases including 5G, massive IoT and AI. It is cloud-native, and powered by the company’s Cuda Tensor Core GPU, which can process 15 teraflops of data per second and up to 140 simultaneous high definition video streams.

“All of this processing basically translates to one single node that is equivalent to hundreds of nodes in a data center,” CEO Jensen Huang said during his keynote speech at last fall’s Mobile World Congress Los Angeles.

Among the partners announced for EGX are, in addition to Ericsson on the vRAN side, Red Hat to help implement Kubernetes container orchestration and deliver a carrier-grade software stack; and Microsoft to integrate its Azure cloud platform to help with AI computation and other intensive workloads, distributed from edge to cloud as required.

Huang described how Nvidia sees 5G and edge computing as being inextricable – twin enablers of many applications that need to process huge amounts of data, often with very quick response times, including AI analytics, cloud gaming, mixed reality, the IoT and software-defined networking (SDN).

He added: “The future is software-defined, and these low latency applications that have to be delivered at the edge can now be provisioned at the edge. That future will become software-defined high performance computing.”

While Nvidia was focusing on use cases, such as factory robotics, which require a combination of 5G connectivity, edge computing and AI, Ericsson was emphasizing how a GPU architecture could make RAN functions themselves far more efficient and scalable.

Noren chose DSS as his example of an intensive, time-sensitive RAN workload which would require very high performance chips before it could be virtualized on cloud infrastructure. DSS allows an operator to run 4G and 5G flexibly on the same band and dynamically allocate spectrum between the two. DSS reschedules spectrum every millisecond, which makes it very challenging for the hardware to handle all the time and frequency variables.

So far, such workloads require special-purpose, fully optimized processors, argued Noren. He said: “We have a multicore baseband architecture that allows us to do parallel processing; we have the capacity to introduce 5G and DSS on the same baseband. Our purpose-built baseband unit is the fastest, most power-efficient unit in the industry. That’s why we can do DSS.”

The same goes for the distributed unit (DU) within a disaggregated vRAN. Most vRAN architectures will run some network functions on a centralized server, while others will be distributed closer to the cell site on a distributed unit, often because they require lower latency.

Ericsson already has roadmaps to virtualize the centralized unit (CU) on off-the-shelf hardware, and has worked on this with Intel. But while the CU requires a great deal of horsepower, the DU brings different challenges because many units will be deployed, so cost and power consumption must be minimized, even while supporting very low latency response and potentially high bandwidth.

Noren said: “If you look at all the available technologies, they are too expensive, too power hungry and too big to be effective compared to our purpose-built hardware for DU. Nvidia has a platform and development framework that we can potentially use.”