Virtual reality (VR,) augmented reality (AR) and artificial intelligence (AI) will dramatically change how smartphones are used. These represent a platform shift, as dramatic as the shift from web to mobile that needs to be powered with new hardware.
These new platforms represent exciting new applications of mobile technology. They are also a possible remedy for overall slowing smartphone unit shipments, as well as Apple’s first year-over-year drop in iPhone shipments and their forecasted continued decline. Without these new platforms, consumers will have little reason to continue to spend as much on mobile tech. There are few reasons to upgrade if the user runs the standard portfolio of apps when a $150 Motorola G4 Play runs them imperceptibly differently than a $700 iPhone.
The issue affects the entire mobile industry because individual app developers, except for game developers, make certain that their apps run fast and power efficiently on older phones as well as new ones because they do not want their apps to be removed by customers who own older phones.
How AR, VR and AI will affect mobile shipments
The faster and more power-efficient Snapdragon 835 SoC from Qualcomm will accelerate the adoption high-performance and upgrade-precipitating AR, VR and AI apps that could reverse slowing mobile shipments. The 835 SoC, with the die size reduced to 10-nm from 14-nm, produced a 27 percent performance increase and a 40 percent decrease in power consumption that will improve the performance and increase the intelligence of AI apps such as Google’s Assistant, voice recognition and translation.
AI has arrived on smartphones. Earlier this week, Facebook demonstrated a Style Transfer app using AI that is akin to Instagram photo filters applied in real time to streamed videos on an iPhone 7. It’s the first AI app to run on a smartphone. The styles are applied to videos using a machine learning (ML) inference model built with Caffe2Go running on the phone. That would not have been possible if the inference model ran on powerful cloud infrastructure. The app would not have performed as well on older phones with slower SoCs. Going forward, building richer apps on larger neural networks on mobile devices will require more powerful hardware, like the Snapdragon 835 SoC.
The first VR platforms, Oculus Rift announced in 2012 and the HTC Vive announced in early 2015, run on expensive PC hardware pumped up on GPU steroids because mobile hardware just was not fast enough at the time to introduce a VR platform. If VR is to become a platform that can be monetized, it has to attract hundreds of millions of users, which pre-supposes that VR has to have the affordability of mobile devices.
Google’s Daydream VR platform is a good case in point. When the Daydream software was first released to software developers, the only smartphone supported by Google to run the software was the Nexus 6P with its Snapdragon 810 SoC fabricated with 20-nm technology.
Google’s support came with a warning that the heat dissipation characteristics of the 6P were inappropriate for a consumer to use as a VR device. And running DayDream drained the 6P battery quickly.
Google’s Daydream-supported device announced last month, the Pixel with its Snapdragon 821 SoC fabricated with 14-nm technology, is much better. Running Daydream, the Pixel gets warm, but not hot, and the better-than-a-day battery performance drops to four hours.
Oculus’ demonstration of its Santa Cruz mobile VR headset at Oculus Connect last month confirms that the company believes mobile is ready to become a full-fledged VR platform. It supports six degrees of freedom (6DoF) using inside-out tracking that maps and orients a user’s movements in the 3D space (reality) to the VR experience in the headset, like what the Oculus Rift and HTC Vive do with their powerful PC platforms.
Snapdragon 835 SoC faster, better performing
At 10-nm, the Snapdragon 835 SoC will have much better performance. An SoC is made up of layers of abstraction. At the very bottom layer are transistors, on-off switches that have two states: 1 or 0, that execute software by changing the states with each clock cycle. The switches are smaller and closer together. Between each cycle, the switches are charged with electrical energy to enable them to switch during the next clock cycle. Because they are smaller, the switches need less energy to switch states. Imagine a 3-foot tall electric light switch. The big switch would take a lot more effort or energy to switch on than a smaller one.
The 835 SoC runs faster because the interconnections to the switches are shorter. Imagine the electrons flowing to the switches to be bowling balls and the interconnects are bowling lanes. If the standard 60-foot lane were cut in half, like the length of the the interconnects were cut in half between the 20-nm 810 SoC and the 835 SoC, it would cut the travel time for the bowling balls to reach the pins in half and would take half the energy to throw the ball. Less power would be lost, too, as the electron travels down the interconnect, like it is less likely the bowling ball would reach the gutter rolling down the shorter lane, reducing power consumption and heat disapation.
Recharging phones will also improve with the 835 SoC’s Quickcharge 4 if it is an equivelent improvement compared to 821 SoC’s Quickcharge 3 that recharges the Google’s Pixel to 80 percent in about 15 minutes.
Upgrade-precipitating AR, VR and AI apps cannot happen soon enough for the mobile industry because they could the revive growth in the high-margin device category. But faster hardware is needed. On spec, the Snapdragon 835 SoC will enable richer apps to be built, advancing the platform shift.