Intel’s CPU Advancements Reinforce Why Architecture-Agnostic Software Is Necessary for Open RAN

The Open Radio Access Network (RAN) has been an important area of focus for development for operators. Over the next few years, ABI Research expects the market to create more than US$24 billion in revenue opportunity by 2027 for vendors. While Original Equipment Manufacturers (OEMs) such as Nokia, Huawei, and Ericsson dominate the market currently, other vendors are trying to further their position in the market, with Samsung and Dell collaborating with Intel to expand their offerings for Communication Service Providers (CSPs).

Samsung recently announced in August 2023 that it is expanding its collaboration with Intel through a new product innovation agreement to allow the companies to continue to advance Virtualized RAN (vRAN). Part of this agreement allows Samsung to integrate its vRAN 3.0 software with 4th Gen Intel Xeon Scalable processors with Intel vRAN Boost to enhance its vRAN technology. The two companies have been collaborating on vRAN technology since 2017, previously using Intel’s previous generation processor; therefore, this extension of their collaboration enables supporting more cells with the same number of servers, increasing power savings and cost efficiency for network operators.

In addition, Dell Technologies has also integrated Intel’s 4th Gen Xeon Scalable processors into its PowerEdge XR 5610 and PowerEdge XR 8000 servers. The main goal of this collaboration is to help operators quickly deploy high-performance and cost-effective radio systems, both virtualized and compliant with Open RAN standards. Dell’s servers have been adopted as Cloud RAN reference platforms by Samsung and Ericsson. The Intel processors integrate Layer 1 (L1) acceleration directly into the System-on-Chip (SoC), enhancing performance and reducing power consumption.

The expanded collaborations with Intel represent further proof of the direction of the Open RAN market. Development in End-to-End (E2E) virtualization of networks is a critical cog in the Open RAN machine, and Intel’s integration of L1 acceleration into the SoC is another step in moving away from solutions that involve custom hardware for Open RAN purposes, and toward the use of off-the-shelf hardware for more scalable and cost-effective solutions. The continued progress in this area has significant advantages for the future of Open RAN, with potential for an accelerated growth phase for the market. Integration of L1 acceleration inside Central Processing Units (CPUs), such as Intel’s latest processor, will play an important role in the long-term investment of vRAN solutions:

• Accelerated Network Evolution: As software upgrades are easier and have marginal costs compared to hardware upgrades, operators can reconfigure their existing networks to support new RAN technologies, as well as add features that reduce operational costs or enable new business models.
• Scalability: Futureproofed software compatibility enables the reuse of existing software, as the number of vRAN users grows and operators must upgrade hardware as capacity is reached. Furthermore, software packages and tools can be configured for a broad range of deployments, ranging from Massive Multiple Input, Multiple Output (MIMO) to small cells.
• Reduced Vendor Dependency: A virtualized L1 layer bypasses proprietary architectures for silicon offerings and non-standardized interfaces, so multiple vendors for hardware and software can be chosen based on an operators’ criteria.

While 2023 has seen considerable attention surrounding Graphical Processing Unit (GPU) innovations like NVIDIA’s Grace Hopper superchip, the maturity of CPU technology for the vRAN use case ensures it outperforms the unrefined GPU technology. Vendors like Mavenir and Fujitsu are exploring GPU-based solutions, but these are still in the experimental phase. Samsung’s dual collaborations—first with Red Hat using NVIDIA’s GPUs and now with Intel—indicate an industry-wide evaluation of both approaches for E2E virtualization. This suggests that due to the novelty of GPU approaches for Open RAN, there is still some uncertainty regarding the long-term validity of switching to GPU-based solutions.

Intel’s latest processor’s integration of L1 acceleration for vRAN is proof of how much more optimized and mature CPU technology is for the Open RAN use case, something NVIDIA has indicated it aims to achieve with GPU technology over the coming years. As further virtualization occurs, the short- and middle-term focus for software vendors should be the development of well-designed, architecture-agnostic software and algorithms to ensure the futureproofing of solutions, regardless of which type of processor operators are using. Below are some strategies that software vendors can use to ensure that future solutions maintain their architecture-agnostic characteristics:

• Cross-Platform Library Utilization: Open RAN solutions must be highly adaptable; therefore, using well-supported, cross-platform libraries ensures that code is portable across different types of hardware, which increases cost and time efficiency and mitigates the need to develop time-consuming, expensive architecture-specific code. Furthermore, the use of well-supported, standardized libraries can make code bases easier to maintain and update.
• Cross-Compilation: Providing an effective method for creating software that can run on multiple types of hardware without requiring multiple development environments is important for architecture-agnostic software. Cross-compilations allow code that is compiled on one platform to run on another, without needing physical access to those architectures. This improves resource efficiency, as the build process can be centralized, enabling quicker compilation and testing cycles, as well as providing a streamlined deployment process.
• Dynamic Configuration: The ability to adapt software to different hardware resources is essential for Open RAN. Dynamic configuration allows system settings or features to be changed at runtime without requiring a restart or redeployment of the software, enabling both automatic and manual tuning of system performance according to real-time metrics and hardware capabilities. This improves ease of management as changes can be made without system interruptions.
• Standardization: Ensuring all software developed follows the industry-standard specifications and guidelines by the Open RAN Alliance ensures that components from different vendors or platforms work seamlessly and makes it easier to add or change components without compatibility issues being a factor. Furthermore, standard interfaces and protocols simplify the design, deployment, and maintenance of systems, accelerating time to market for solutions, and helps in adapting to future technologies without significant overhauls.

The integration of L1 acceleration is not the only evidence of CPU market maturity and optimization compared to GPUs. Intel’s FlexRAN reference architecture introduced the disaggregation of RAN software and hardware, enabling vendors to virtualize Distributed Unit (DU) and Centralized Unit (CU) functions and host Layer 2 and Layer 3 functions on Commercial-Off-the-Shelf (COTS) hardware using Intel processors. The reference architecture allows for time to market acceleration, helps avoid vendor lock-in, and assists in optimizing energy consumption. FlexRAN is widely established and very mature in the industry, whereas NVIDIA is just starting on this path. As the Open RAN market grows, the focus for processor vendors should be on following the same strategy to succeed.