Nvidia, Dominating Autonomous Technology

| About: NVIDIA Corporation (NVDA)


Nvidia seemingly came out of left field in the autonomous wars currently being fought, landing a contract with the ultimate autonomous car company, Tesla.

How did a company, like Mobileye, whose sole competency is autonomous car technology, get its lunch eaten by what was once just a GPU company.

GPUs have a lot of advantages when dealing with the problem sets of autonomous vehicles versus CPU technology solutions.

We'll show why Nvidia's domination in this space is likely to continue.

To understand Nvidia's (NASDAQ:NVDA) recent remarkable entrance into the autonomous vehicle race that has brought into question Mobileye's (NYSE:MBLY) position as the top dog, we'll have to go into the technology. So let's go into the tech a bit deeper - you'll be a pro the next time this topic of conversation comes up at your holiday cocktail parties. And I promise you, the conversation about autonomous vehicles isn't going anywhere.

Nvidia made its mark in the GPU, or Graphics Processing Unit, space when it popularized the term back in 1999. Since then, the company has seen its prospects seemingly rise and fall with the tide. First, computers needed to have GPUs to run the games people started playing on PCs. Then, with the advent of console gaming systems, GPUs took a back seat. There was a gaming renaissance in the 2005-2008 period where GPUs became in vogue again, only to fall out out of fashion when companies sold their onboard graphics solutions as sufficient for most consumers. Now, as consumers stand ready to adopt VR and tech fans who are flush with cash from their high-paying Silicon Valley jobs want the latest and greatest in processing power, Nvidia-supplied GPUs are in demand once more. Enough with the history of Nvidia, let's get down to business.

Why are GPUs good at handling the challenges and data sets that autonomous cars present us? Well, GPUs have lots of little processors versus CPUs with 2, 4, 8 and sometimes 16 separate processors. Some GPUs have thousands of what are called stream processors to process data, albeit at a slower rate on their own than CPUs can handle. Because all these processors run in parallel, i.e., at the same time, they can process relative simple tasks, but tasks that have massive amounts of data. CPUs can handle a big, long, complex task with ease, something with which a GPU would have a hard time.

When things are simple, but there are a whole lot of them, GPUs really hold their own. Does this start to sound familiar? Think about driving? Is it that complex? No, not to me at least. Can you drive on a rather instinctual basis? I sure can, I don't even think anymore. When you are driving, are you processing loads of information? That little old lady crossing the street, while there is some punk to the left of you that you know is about to cut you off? You know what I mean here. Driving is a simple task, but with loads of information flying at you. If your brain were a computer, while it was driving, which processor would you rather use? You'd rather have a powerful GPU.

The Nvidia chips that are used in their autonomous package use AI deep learning technology that is based on neural networks copied from the human brain itself. The technology looks to replicate how the human brain makes decisions and learns the best way to make these decisions. Because the tech uses a hierarchy, the system separates complex problems into lots of little ones. Enough tech mumbo jumbo, think about how you might identify a aircraft landing at an airport. First, you recognize the shape of the aircraft that you were taught as a kid. Ok, this is an airplane. Then you look for further clues. Is it white or dark? If it is white, your brain likely tells you it's a passenger aircraft and not military. Is it big? Does it have 4 or 2 engines? If the aircraft is both big and has 4 engines, you probably recognize it's either a 747 or an A380/40. Does it have a hump at the front? Sure does. Right away you recognize this aircraft is a 747 - your brain goes through this hierarchy almost instantaneously. But was this hard to get to? Not really, you just had a binomial tree in your brain that got you there - you just did it super fast.

That's all that is happening when you are driving, with some different functions. You see an octagon on a post up ahead, which looks red to you. You ascertain this is a stop sign. You know from previous experiences around stop signs that there are more dangers, more pedestrians, more cars, so you instinctively slow down while you come up to the sign (plus it's the law - small detail). These are exactly the type of processes a GPU excels at handling. The computer recognizes a stop sign ahead; they assign a higher danger rating to this, so the computer knows to slow down. In addition to this, the computer knows there are often more pedestrians around stop signs, so it takes more time scanning and identifying them. Driving is just a whole bunch of little tasks over and over again. It's perfect for GPU processors. I hope that wasn't too painful - just think you'll now know more about something that actually matters than that guy at cocktail parties who is always showing off their new iPhone. So what does Nvidia's driverless car solution look like?

Nvidia DGX-1 & PX, PX 2 and PX 3

Behold the box pictured below - that is a brain:

DGX Click to enlarge

(Source: Nvidia.com)

It doesn't look like much, does it? That's because you aren't going to see this box. It's meant to be tucked away out of sight. Nvidia is calling this the "World's Supercomputer in a Box," and who am I to try and poke holes in such a claim. What's installed in cars won't be this box, but just a board with chips on it. This box is what made it all possible. The new Model-S with hardware 2.0 (8 cameras, 12 ultrasound sensors, and forward radar systems) currently being showed off in videos by Tesla (NASDAQ:TSLA) use the same neural networks with their PX 2 system as those on which the GDX-1 was built.

I'm going to try and make a very complex situation easily understandable again. Recent studies have indicated driverless cars will need to process, and thus create, 1 GB/s or 1 Gigabyte of data per second. The PX 2 is advertised with the ability to handle 70 GB/s, which would exceed what's needed on pure Level 4 Autonomy (see the 5 levels here). The PX 2 is capable of 24 trillion operations a second - that's more than 3X the population of planet Earth in operations a second. However, just because a system can handle the throughput doesn't mean the sensors can capture that much data - at the current time, they can't. Nothing works quite as well as advertised, and there are inevitable bottlenecks throughout the whole system.

The PX 2 is the only system currently on the market capable of handling what I would call real autonomous driving; most of the cars in the videos that have gone viral have had these systems installed (see the videos here). The PX 3 will be an even better system, and will undoubtedly lower the cost of the PX 2. What does all this mean in English? There is a computer system available today that can handle level 4 autonomous vehicles, and it is currently made by Nvidia. After you read that last sentence, I think the shares' 180% rise this year makes more sense. So where does Nvidia go from here?

Nvidia, Autonomous Applications, and the Competition

So if Nvidia can remain the dominant player - which it JUST became - in this space going forward, how are the company's prospects related to this vertical? Pretty darn good. Autonomous cars, in my opinion, are one of the most important technologies to come around since the internet itself. If you have a loved one sick and receiving treatment, I'm sure you think otherwise - advancements in medical technology have been amazing. Mostly though, people deal with the transportation of their selves, families, and goods they consume more in aggregate than the health of their selves and families.

This year, an estimated 75 million cars were sold globally. Nvidia's current offering costs $1500. If 10 years into the future 100% of cars have autonomous functions, Nvidia cuts that price point by 1/3rd and commands 15% of the market, assuming the market grows by just 1% in those 10 years, the company's sales in the vertical could exceed $6 billion. That would be nearly double its 2016 revenue. Anyone who has looked at a chart of NVDA recently is aware that the market keenly understands this possibility. NVDA shares are among the best performers in 2016.

I don't want this to get more complex than it already is, but I believe autonomous cars are going to be nearly as important as the internet itself. Nvidia has a tremendous amount of runway ahead of it. Who is the competition? We already brought up Mobileye, but the company is in the process of getting its lunch eaten by Nvidia. Delphi (NYSE:DLPH) is super-active in the space - its V2E (vehicles-to-everything) system connects and communicates with, well, everything. Even people, if they have the right chip in their smartphones, will be able to communicate with oncoming cars - even if they aren't aware that they are. But Delphi's offering as of now does not represent a complete solution.

Intel (NASDAQ:INTC) is trying to throw its hat in the race, but currently, it's no contest. In fact, Intel's obsession with telling consumers what they don't need - in this case GPUs, because according to Gregory Bryant VP of desktop, "Integrated graphics have nearly caught up" - might have hurt the company's chances. We've already seen why GPUs are more up to the task of autonomous driving that their CPU brethren. Intel might have a made a miscalculation here, one of many lately.

Whose left? AMD makes ATI GPUs - could those be used on autonomous vehicles? They could, but AMD hasn't given any indication it is going down that road yet. The company is quite small; still, if it decided to build a solution, and even one automaker went with it, AMD could move the needle. So, for the time being, Nvidia is the winner. The automakers have voted with their feet; BMW (OTCPK:BMWYY), Tesla, Honda (NYSE:HMC) and Volkswagen (OTCPK:VLKAY) have all bet on Nvidia's offering.


Understanding the inner workings of all these boxes that are "smart" or "connected" can be a mind-numbing exercise. At the very least, I hope you came away with a deeper understanding of what is actually happening from a processing standpoint when a driverless car drives, and why GPUs are superior to tackle these issues. The data flowing at you when you are driving is immense - and GPUs love these data sets. Many analysts were scratching their heads when Tesla dropped what seemed like the powerhouse, Mobileye, and went with Nvidia, which many viewed as just a company that makes games cooler to play. Nvidia is keenly aware of the advantages of GPUs over CPUs; the company has been waiting for just this moment. Its solutions have further reaching applications than just autonomous vehicles. The intelligent bots you will be communicating with in the future will be driven with this tech as well, and Nvidia will be at the forefront of these technologies. Expect to be hearing about this company a great deal in the future.

Disclosure: I/we have no positions in any stocks mentioned, but may initiate a long position in NVDA, AMD over the next 72 hours.

I wrote this article myself, and it expresses my own opinions. I am not receiving compensation for it (other than from Seeking Alpha). I have no business relationship with any company whose stock is mentioned in this article.