CPU, GPU or … VPU?

Winners, losers and observations from CES .


Where is the semiconductor industry going in the post-smartphone era? What trends are going to shape next-generation applications and SoC development?

Just by walking around the CES show floor this year, I would say advanced visual processing technology is the horse to put money on. It was everywhere, from ADAS systems, drones, to GoPro cameras, IP cameras with embedded facial recognition, motion detectors, virtual reality, augmented reality, displays and a whole lot more.

Source: Author photo

The smartphone wars are over. And your car is your new iPhone if all the technologies demonstrated from CES come into fruition. This is classic technology evolution, not revolution.

Figure 1. The evolution of computing, from Babbage’s analytical engine, to the JOHNNIAC mainframe, IBM PC-AT, Apple iPhone, and now your new car. (Sources: Computer History Museum)

Automotive: Learning from aviation
Cars were everywhere at CES, but there are huge challenges to integrating consumer technologies into these platforms. User interface technology in automotive is the biggest hurdle here. The most promising technologies stem from aviation, including heads-up displays (HUD) and voice prompts. Visteon demonstrated a windshield HUD system that overlaid key information (like speed), supplementary information (like a map), and infrared (IR) video for obstacle avoidance. It was the slickest automobile glass cockpit I have ever seen.

Voice recognition has a long way to go based on what I saw (and heard) at CES, and driver control innovation is stagnant. I don’t expect to see a HOTAS setup in a car anytime soon.

Figure 2. A great display of automotive electronics system integration by Hyundai Mobis. (Source: Author photo)

A sore point for the human factors engineer in me is that many OEMs were offering touch screens in the car to adjust the infotainment/navigation system. This is a step backward. Touchscreens are just not as safe has having old fashioned dials because you have to look away from what’s going on outside the vehicle to make adjustments in the vehicle. That’s why aircraft cockpits have buttons, knobs and dials rather than touchscreens.

Display technology: Thin is in, and so is OLED
LG displays reigned supreme at this year’s show, with OLED models not much thicker than my iPad. If you haven’t seen an OLED display, you need to. It blows plasma and LED TVs out of the water. Screen-on-glass technology has come into its own, the most impressive example being Samsung’s transparent displays which will be great for advertising and retail. TVs are getting larger in both size and resolution, with quite a few large 8K TVs on display. These aren’t much use for consumers because of the lack of 8K content, but they will be useful commercial applications.

Figure 3. LG Electronics’ booth was packed with people gawking at their new OLED TVs.

Virtual reality: Really annoying
Many of the virtual reality-type displays at CES indicate this technology is still in its infancy. Oculus Rift was omnipresent, but there were a ton of competing companies and that could spell commoditization very quickly.

In any case, nearly a quarter of the booths had some kind of VR goggle setup as a gimmick. One VR standout was Sleep Number, the inflatable bed company, that had a booth with just carpeting and attendees sporting VR goggles. I don’t know what they were showing, but there weren’t any beds in there.

My take is that “virtual reality” as we know it today won’t be as useful as augmented reality, like the automobile HUD. You got speed readouts, information on weather and a warning when my kid chases his soccer ball right in front of your car. Theses HUDs allow you to see graphic representations what’s in front, behind and beside you so can maintain much better situational awareness. Bottom line: Ugly googles on your face are not going to change your life, but adding this technology to existing products to solve real-world problems will.

IoT is dead: Long live IoT
In addition to smart phones, the downtrend seems to encompass the so-called Internet of Things. Most IoT solutions seemed to be evolutionary such as adding moderately useless features into significantly overpriced kitchen appliances. I think the buzz around IoT has faded quickly because a killer app hasn’t emerged.

My favorite expensive solution in search of a problem was the IoT baby car seat. I don’t think we will see that thing on display again next year.

Advanced visual processing: Rise of the VPU

Everything is evolving, but the technology that first appeared in smartphones and other consumer devices is now going everywhere. This is especially true in advanced visual processing. Whether in our cars, on our TVs or somewhere else in front of our faces, expensive video processing using cheap cameras and displays will change the products we see in the future. At CES I saw lots of block diagrams with “visual processing unit (VPU)” as the key differentiating IP. Combine these VPU-heavy systems-on-chip with machine learning and we are on the way to autonomous systems that can be trained to function without constant human intervention. I don’t expect self-driving cars to be ubiquitous any time soon, but I do expect these technologies to help make new cars much safer than they are today.

SoC design and the future
One question I have is how well our “smartphone-centric” industry will adapt to these new markets. Will semiconductor companies try to sell their existing mobile architectures into these new areas, which I think is like trying to hammer a nail with a screwdriver? Or will our industry innovate and lead a new technology revolution with semiconductor companies at the forefront? Companies like Google, Ford, Apple and Tesla are not waiting for us to figure this out.


realjjj says:

If you talk about the future, then it’s glasses centric not smartphone centric and that changes a lot .It also makes visual processing even more important and difficult.
All those screens in cars,redundant. So are TVs and phones and so on.
Lets see what hits us at MWC, with some luck maybe we Leap 5 years into the Magic future – just speculating, don’t have reliable info.

Kurt Shuler says:

I got lasik so I could ditch my glasses, so I’ll never sign up to wear those suckers again unless I have to.

Nirvana for me is being able to enter my car, have my collection of tunes start playing automatically, have my destination shown in a familiar Waze map on my windshield HUD, and pay more attention to what’s outside the car than all the tech inside. It will be interesting to see what the new Apple CarPlay and Android Auto integrations can enable here.

realjjj says:

Glasses offer an always on screen , a cost benefit vs big screens (huge cost benefit when you factor in volumes once they ramp ), a much bigger display area vs phones , opportunity for lots of sensors and more. Aiming straight for contact lenses or beyond would be too optimistic.
You seem to focus on cars, i’m thinking CE in general. In cars i like the idea of biometrics to identify the driver and adjust settings, would help with your goal too – cars are the only area where i do like biometrics, elsewhere it is convenient but a password we can’t change or conceal is far from ideal.
The future of cars seems tricky. On one hand CaaS with autonomous is a huge deal and offers all kind of new options but in places where public transport is very popular such services could make traffic worse by taking customers away from public transport. At the same time we do need to reclaim the space we lost to cars while electric scooters and bellow (segways and less) seem to be about to boom. Balancing all of this will be complicated .

Back to glasses, i hinted at Magic Leap since they just got 800 million in new funding and a device showed up in a benchmark so maybe they have hardware soon. Their latest video on their Youtube channel made me wonder if they might be going after education and play first. It’s easier on the software side to go after a niche and targeting all kids in the developed world first would be a big enough market to get things rolling.It’s hard to get everything to good enough when it comes to glasses so not hoping for too much but sooner or later someone will get there.
Glasses can help in so many many ways. For media you have 360, imagine watching a Tarantino car chance and being in the car able to look around . Then there are things like gardening, cooking, ikea or Lego manual , driving or even micro soldering- you use the cam to zoom in. They could warn you of an incoming threat while walking on the sidewalk or show you the path to escape an avalanche while skiing. For work you could combine virtual and physical hardware with unlimited options and great flexibility. Google Now like features and an intelligent UI can assist you with information in any context. Always on is something no other device offers and a tiny screen should end up having a big cost advantage over a big one once the tech matures. Wearing glasses 24/7 is not great and wearing heavy glasses 24/7 must be much worse. The hardware will be hard,, volume and weight are huge complications.


I think that both glasses centric and car centric approaches are equally interesting and both of them will develop into new apps.
SO far, having GPGPU consiting of 20 servers with K80
(4992 GPUs in each K80) – we develop also interesting apps

Leave a Reply

(Note: This name will be displayed publicly)