Quectel Engages in Valuable New 5G Edge AI Opportunities with NVIDIA

Quectel and NVIDIA are working together to integrate Quectel’s RM500Q 5G cellular communications module with NVIDIA’s Jetson AGX Orin Artificial Intelligence (AI) Graphics Processing Unit (GPU) module. Both products are powerful, miniaturized, and specialized components that Original Equipment Manufacturers (OEMs) can use to build their own devices. So far, the RM500Q and Jetson AGX Orin have been combined within an edge AI gateway manufactured by Suzhou-based Chinese firm TZTEK. The hope is that this pilot implementation will act as a blueprint for the common use of Quectel and NVIDIA’s products in other OEMs’ equipment, having proven how well suited they are to each other, and the mutual benefits each conveys upon the other, purporting the oft-cited collaborative advantages of simpler development, less time and cost spent, and a faster time-to-market.

The Jetson AGX Orin comes in 64 GB and 32 GB configurations and has configurable power consumption of between 15 W and 60 W. The two different Random Access Memory (RAM) configurations of this GPU product both use NVIDIA’s Ampere architecture, but feature a different number of Compute Unified Device Architecture (CUDA) cores and Tensor Cores, hence, a different maximum GPU frequency of 939 MHz and 1.3 GHz, respectively. The more powerful 64 GB module has a maximum AI performance of 275 Tera (i.e., trillion) Operations Per Second (TOPS). The primary intended use case for the Jetson AGX Orin is, as stated by NVIDIA, in robotics and other autonomous machines. The 32 GB Jetson AGX Orin module will be commercially available in July 2022, with its 64 GB bigger brother going on sale in October 2022, for US$899 and US$1,599 per unit.

By comparison, Quectel’s Non-Standalone (NSA), and Standalone (SA) sub-6 GHz RM500Q series of modules are a fragment of that price at approximately US$200 each, although prices will vary by volume and by far more than the Jetson AGX Orin. The RM500Q uses the popular M.2 form factor Printed Circuit Board (PCB) connector. This form factor is common in around two-thirds of all 5G IoT module models and is useful for flexibly allowing various components to be modularly integrated, accommodating double-length and double-sided components in the same PCB footprint. The RM500Q is cited as an industrial-grade module, backward compatible, and designed to encourage OEMs to eventually migrate from Long Term Evolution (LTE) to 5G. It has a maximum downlink speed of 2.5 Gbps and uplink of 900 Mbps, plus Global Navigation Satellite System (GNSS) support for Global Positioning System (GPS), GLONASS, BDS, and Galileo. It is Release 15 complaint and based on Qualcomm’s Snapdragon X55 modem chipset.

5G is taking time to find its feet within the Internet of Things (IoT). The IoT thrives on the reuse of technology that is already fully paid for—developed, deployed, and commoditized by the consumer mobile broadband market. But 5G is taking time to prove its differentiated credentials there, too. Gateways and routers have been a natural initial 5G application. Gateways and routers are largely stationary, so their rollout is coordinated with assured 5G coverage. And their role as points of aggregation for the backhaul of data makes them strong potential beneficiaries of the greater throughput rates and lower latency that 5G offers. 5G fixed wireless terminals are a market that Quectel—and all cellular module vendors—are actively targeting. Quectel’s 5G module shipments kicked off in earnest in mid-2021, following a tender win with China Mobile for the supply of 150,000 units.

Edge computing in the IoT can serve two purposes. It can selectively process and filter information collected from local sensor arrays to minimize the amount of data needing to be sent to backend systems, where algorithms then calculate actionable insights. Or it can bring that same cloud intelligence to the edge, distributing previously centralized processing power to minimize the time taken for decisions to be made locally within the context of the entire network. The first example is designed to only require modest connectivity, while the second deliberately makes use of as much bandwidth as is available. Quectel is anticipating its AI edge computing relationship with NVIDIA to serve “advanced robotics, unmanned ground vehicles, low-speed automated driving systems, and intelligent transportation,” all of which are low volume, but very high unit value deployment scenarios.

The next step for Quectel is integrating its Release 16 modules with NVIDIA’s products to provide Millimeter Wave (mmWave), Ultra-Reliable Low-Latency Communications (URLLC), network slicing, and a 5G Local Area Network (LAN). If sub-6 GHz has been cutting edge, and slow to pick up pace in the IoT, mmWave is considered bleeding edge in an IoT context. The only known meaningful commercial shipment of mmWave IoT modules during 2021 was by Sierra Wireless within its own-brand range of high-end XR-series gateways. Anything that an IoT module vendor can do to acquire a flagship logo like NVIDIA for promoting its newest and most expensive product lines is worthwhile. NVIDIA’s name has long been synonymous with high invention and innovation. It owns hugely valuable Intellectual Property (IP), after happily finding the future of AI land in its lap, courtesy of NVIDIA pioneering parallel processing work originally undertaken for the sake of video game graphics.

Quectel has careened its way to the most eye-opening sales figures in the IoT module market, seemingly in record time, posting a turnover of more than US$1.6 billion in 2021. Not all of this revenue was from cellular modules, and not all of Quectel’s cellular modules were for use in IoT applications, although, in both cases, the majority of it was. Despite the IoT being Quectel’s original and primary market of focus, Quectel is a manufacturer that is keen to see its products find their way into as many applications as possible, regardless of how analysts choose to categorize them. AI is touted as being applicable to both the consumer mobile broadband market and the IoT, and can even be considered a point of overlap between man and machine, with machines trained by man capable of then acting on mans’ behalf.

Immediately preceding Quectel’s May 24 announcement, on May 23 at COMPUTEX in Taiwan, NVIDIA stated that more than 30 device OEM partners had been confirmed as announcing the commercial availability of Jetson AGX Orin-based devices and systems. NVIDIA cited the product examples of “servers, edge appliances, industrial PCs, carrier boards, [and] AI software” for both “commercial or ruggedized applications in robotics, manufacturing, retail, transportation, smart cities, [and] healthcare” as forthcoming. However, the number of those partners and products that would need 5G connectivity and that may have chosen to go with Quectel as a module supplier was not discussed, aside from the aforementioned test bed integration by TZTEK.

It should be noted that NVIDIA also boasts Telit as a Jetson Orin partner. So, just because Quectel has been successfully involved in one early Jetson AGX Orin integration project does not mean that others cannot be. AI device OEMs are likely to want a choice of prospective 5G hardware partners and may well choose their own existing preferred supplier. The market for AI edge computing will demand the highest performance and reliability and only a handful of the world’s 50+ cellular IoT module vendors will have the right reputation to be chosen to serve. It can help to be in the game early, as Quectel’s balance sheet exemplifies that nothing succeeds like success. Big potential rewards come with big investments and risks; however, the opportunity for 5G in edge computing seems real enough, judging by NVIDIA’s growing list of Jetson AGX Orin partners.