At a Monday press event that took place at the CES trade show, Qualcomm unveiled a family of ADAS and autonomous driving system-on-chips (SoCs) that will be sold under the Snapdragon Ride Safety label. The company also unveiled accelerator hardware that leverages its new silicon, and a comprehensive software stack meant to run on top of it.
Qualcomm says its ADAS/autonomous platform, broadly referred to as Snapdragon Ride, will be "available for pre-development to automakers and tier-1 suppliers in the first half of 2020." The company said it anticipates production vehicles featuring its platform will be on the road in 2023.
Depending on an automaker's needs, a Snapdragon Ride solution can be tailored for one of three use cases: Basic Level 1/Level 2 ADAS for functions such as emergency braking and lane-departure warnings; Level 2+ solutions that can automate activities such as parking and highway driving; and fully-fledged Level 4/Level 5 autonomous vehicles.
The processing power delivered by Snapdragon Ride for AI/deep learning operations ranges from 30 tera operations per second (TOPS) for L1/L2 ADAS to a massive 700 TOPS for L4/L5 systems. Qualcomm claims its solutions deliver "industry-leading thermal efficiency," and that Snapdragon Ride will consume only 130 watts of power at 700 TOPS.
Not surprisingly, the Snapdragon Ride makes extensive use of the massive chip R&D investments Qualcomm has made over the years in support of its Snapdragon processors for smartphones and other end-markets. CPU core, GPU, image processor and AI co-processor IP previously developed for other Snapdragon chips are all leveraged by the Snapdragon Ride Safety SoCs.
The Snapdragon Ride will be squaring off against Nvidia's Drive autonomous driving hardware/software platform, as well as the ADAS and autonomous driving offerings provided by Intel's Mobileye unit. Both Nvidia and Intel/Mobileye already have numerous autonomous engagements with top-tier automakers and parts suppliers, and Mobileye has long been a top ADAS vision processor supplier.
Nvidia is less than a month removed from unveiling the Drive AGX Orin, an autonomous driving SoC promised to deliver 200 TOPS of performance while consuming less than 70 watts of energy. Orin, which is meant to go into production vehicles by 2022, is much more power-efficient than Nvidia's older, multi-chip, Drive Pegasus hardware, which deliver 320 TOPS of performance but consumes up to 500 watts.
Mobileye, meanwhile, is working with partner STMicroelectronics (STM) to commercialize its EyeQ5 autonomous driving SoC. The EyeQ5 is forecast to enter mass-production by March 2021, and delivers 24 TOPS of performance while consuming 10 watts of power. Mobileye and STMicro have noted that a vehicle could feature multiple EyeQ5 chips.
In addition, Tesla (TSLA) is nine months removed from unveiling a proprietary ADAS/autonomous driving SoC known as the FSD (Full Self-Driving) computer. The company's third-gen Autopilot computer, which features two FSD chips, is said to deliver 144 TOPS of performance while consuming 72 watts of power.
Qualcomm is definitely arriving much later to the autonomous driving processor market than Intel and Nvidia, each of which have made large software investments to go with their silicon R&D. However, recent push-outs for autonomous driving timetables could make it a little easier for Qualcomm to gain an audience with major automakers, at least provided that its impressive performance and power efficiency claims hold up.
Whereas automakers once talked about offering cars supporting Level 4 autonomy by 2020, expectations for the start of this decade have become far more measured. At recent investment conferences, Nvidia execs said they now expect high-volume design wins landed for Level 2+ solutions to start ramping around 2022.