Opinion: Intel focuses on 5G infrastructure
Despite the cancellation of this year’s Mobile World Congress show in Barcelona, quite a few tech companies are still making important announcements that were originally planned for the show. Not surprisingly, several of those announcements are focused on 5G.
One of them—perhaps somewhat surprisingly—comes from chip leader Intel. The company sold its 5G modem business assets to Apple last fall, and many considered that move to be the company’s exit from the 5G world. But Intel has a much bigger, though significantly less well-known, business creating chips that help power the network infrastructure equipment that sits at the heart of today’s 4G LTE and 5G networks, including base stations and other core components.
For years, much of the network silicon inside these devices was custom designed and built by the vendors making the equipment—companies like Ericsson, Nokia, Huawei, etc. However, with the growth of more software-based networking advancements, including things like network function virtualization (NFV), as well as increasing demand for general compute performance to handle applications like AI at the edge, Intel and other specialized vendors like Marvell have seen strong interest in their off-the-shelf “merchant” chips.
To better address those needs, Intel made several different announcements focused on adding the computing power and specialized resources that new 5G networks require at multiple points along the network path. Starting at the network edge, the company introduced a new version of its Atom processor, the P5900, that’s specifically targeted towards wireless base stations. Based on a 10nm process technology, the new SoC (system on chip) integrates a next-generation set of Atom CPU cores along with network acceleration functions that are specifically targeted for radio access functions, small cells, and other edge of the network applications. Given the strong expected market for 5G-focused small cells that millimeter wave, mid-band sub-6 GHz and CBRS-based spectrum demand—as well as the potential to do cloud-based computing workloads on the edge, such as AI—this looks to be a very interesting opportunity.
For more compute-intensive workloads at the core of the network, the company also chose to make a number of additions to its second-generation general-purpose Xeon Scalable server processors as part of this 5G announcement. Facing intensive pricing and performance pressure from AMD’s second generation Epyc server processors, Intel added 18 new SKUs to its lineup that offer more cores, faster clock speeds, and large cache sizes at lower prices than some of its initial second-gen Xeon Scalable parts. In terms of performance, Intel touted up to 56% improvement for NFV workloads versus some of its first-generation Xeon Scalable CPUs (though the company didn’t clarify performance improvements vs. some of the earlier second-generation parts).
Another key element that’s essential to speed up the performance of core telecom markets are programmable chips that can be optimized to run network packet processing and other functions that are critical to guaranteeing lower latency and meeting consistent quality of service requirements. These points are becoming particularly important for 5G, which has promised improved latency as one of its key benefits versus 4G.
FPGAs (Field Programmable Gate Arrays) have traditionally done much of this kind of work in telecom equipment, and Intel has a large, established FPGA business with its Agilex line of chips. The power and flexibility of FPGAs do come with a cost, however, in terms of both pricing and power, so Intel debuted its first all-new design in a chip category it’s calling a structured ASIC and a product that’s currently codenamed Diamond Mesa.
The idea with a structured ASIC is that it’s essentially only partially programmable, and therefore sits between an FPGA and custom-designed ASIC. From a practical perspective, that means it offers faster time to market than building a custom ASIC at a lower price and power requirement than an FPGA. To ease the transition for existing FPGA users, however, Intel has designed Diamond Mesa to be footprint compatible with its FPGAs, making it easier to integrate into existing designs. The real-world benefit is that, used in conjunction with the latest Xeon Scalable CPUs, Diamond Mesa will let telco equipment providers create products that can handle the increased performance, latency, and security demands of 5G networks.
The last portion of the Intel announcement centered on, of all things, a wired ethernet adaptor. While much of the focus for 5G and any other telecom network is typically on wireless technologies, the reality is that much of the infrastructure still uses wired connections for interconnecting different components across the network core and to enable certain capabilities.
Particularly for applications that require time-sensitive networking—including things like precise industrial automation—we’re still several years away from being able to ensure consistent real-time signal delivery over completely wireless networks. As a result, Intel’s new 700 series network adapter—which incorporates hardware-enhanced precision time protocol (PTP) support that leverages GPS clock signals for cross-network service synchronization, according to the company—still has an important, if not terribly exciting, function to fulfill in 5G networks.
All told, the Intel 5G network infrastructure story offers a pretty comprehensive set of offerings that highlight how the company has a bigger role to play in the latest generation wireless network than many people may initially realize. Of course, it’s a big field, with a lot of different opportunities for many different vendors, but there’s no doubt that Intel is serious about making its presence felt in 5G. With these announcements the company has made several important steps in that direction, and it will be interesting to see what the future brings.