NVIDIA isn’t content with dominating the PC gaming market with its GTX 1080,GTX 1070 and GTX 1060. Now the company wants to do the same in the car industry by powering the next generation of autonomous cars.
The good news is that NVIDIA’s latest CPU is quite a capable beast when it comes to processing power. The new NVIDIA CPU has been developed, from the ground up, specifically for autonomous cars.
NVIDIA revealed its highly anticipated DRIVE PX2 supercomputer at CES earlier this year in January.
DRIVE PX2 supercomputer is also for in-car use only. NVIDIA’s DRIVE PX2 is not just an ordinary supercomputer. It is a supercomputer that is a liquid-cooled processor for autonomous cars.
Of course, at the moment there is no way to make sure if NVIDIA’s latest DRIVE PX2 would become a major player as far as computers for autonomous cars are concerned but the DRIVE PX2 does indicate that NVIDIA is slowly and but surely getting serious about providing powerful chips for self-driving cars.
Just last year NVIDIA introduced its DRIVE PX in-car computer which was also a pretty impressive computer. But this year’s DRIVE PX2 supercomputer is another beast altogether.
The DRIVE PX2 has power which is equivalent to 150 MacBook Pros. That insane amount of power is the sole reason why DRIVE PX2 has made such waves in the industry.
NVIDIA’s DRIVE PX2, as a package, has 12 CPU cores and can provide processing power up to 8 teraflops.
Just to put it in perspective, Microsoft’s yet unreleased game-changing revolutionary Project Scorpio is only capable of outputting 6 teraflops.
With 8 teraflops of power at its disposal, NVIDIA DRIVE PX2 can give video cards such as Titan X a good run for its money.
NVIDIA has also stated that the DRIVE PX2 is capable of achieving 24 trillion operations a second. That gigantic amount of potential is expected to be sufficient for powering self-driving autonomous cars.
NVIDIA DRIVE PX2 is without a doubt the most powerful processing unit that the industry has ever seen. Jen-Hsun Huang, who is NVIDIA’s CEO, recently said that NVIDIA DRIVE PX2 was the first supercomputer which was made specifically for self-driving cars.
But readers might legitimately wonder as to why NVIDIA went for such a powerful machine. After all, in a practicality, NVIDIA DRIVE PX2 looks to be an overkill as far as processing power is concerned.
Jen-Hsun Huang answered this questions by saying that DRIVE PX2 specs were necessary for building competent self-driving vehicles.
He also said that since self-driving cars would utilize technologies such as “deep learning”, it was of vital importance that their CPU could process large amounts of data.
Huang was of the opinion that self-driving cars of the future would make heavy use of maps and sensors and no matter how well these were programmed, the car would still need to know how to handle changes in its environment on the fly.
NVIDIA’s CEO continued further and said that self-driving cars of the future should be able to react appropriately if, for example, a child jumped out on the road. He said that deep learning would enable self-driving autonomous cars to teach themselves for each and every unexpected scenario that might take place while on the road.
Of course, all of that would take time but NVIDIA’s CEO looked confident about DRIVE PX2’s capabilities.
Huang also spoke about how he felt that humans won’t be able to realize the full potential of self-driving autonomous cars unless and until they could solve the problem of city driving.
Huang further added, “Bikers are on the same road you are… People are sometimes following the rules and most times not… It’s very chaotic and very hard.”
It is safe to say that with DRIVE PX2, NVIDIA has doubled down on its last year’s vision of facilitating high-end PC gamers to play the latest and the greatest games on the market with the highest settings possible.
NVIDIA GPUs are now powerful enough that not only can they play high-end games with ease but can also provide the necessary processing power to perform the insane levels of computation that is required for self-driving autonomous cars.
NVIDIA has also announced its latest deep learning platform which is called DIGITS. Not only that but the company is also currently testing DIGITS with its own self-driving autonomous cars.
The deep learning initiative is essentially a method for self-driving autonomous cars to save everything they learn while on the road (and out open in the environment) and share that knowledge with a cloud-based network.
That knowledge can then uploaded to other self-driving autonomous cars. NVIDIA DIGITS has been designed to make that process easier for car companies as they kick off the all-important testing phase of self-driving autonomous cars.
NVIDIA’s CEO, Huang, also assured reporters that the company would allow other car manufacturers to own their own deep learning neural networks and that NVIDIA was just helping to get the process started as quickly as possible.
As mentioned before, DIGITS aided NVIDIA immensely in producing its own deep neural network, Drivenet. NVIDIA Drivenet features a total of nine “inception layers.”
Huang told reporters that Drive Net’s nine inception layers were like nine individual neural networks that were embedded within each other.
He also explained that when information was allowed to run through the network even once, the process required 40 billion operations. Huang affirmed that Drivenet needed a lot of computing power behind it in order to perform such massive tasks.
Huang also revealed that NVIDIA’s Drivenet was able to identify a total of five separate classes of objects such as pedestrians, motorcycles, cars, signs etc.
Different objects are color coded by Drivenet and that detail to object recognition, according to Huang, was extremely important for self-driving autonomous cars of the future.
Huang also told reporters that Audi utilized Drivenet study the visual data it had collected from a snowstorm and after just one night, the company was able to detect data that skipped the human eye.
Needless to say that all the computing power in the world is of no consequence if your technology isn’t adopted by the masses.
To address this issue, Huang informed reporters that Volvo was the first company that officially adopted the NVIDIA DRIVE PX2 which, Huang thought, was a pretty decent start.
To further add more value to its products, NVIDIA also showed off the Parker System on a chip which powered DRIVE PX2.
The new processor basically has a mammoth 256-core system and that allows it to provide 1.5 teraflops of power for “deep learning based self-driving AI cockpit system.”
NVIDIA announced these technical specifications through a blog post on its official website. The blog post also revealed that the Parker System could additionally perform 24 trillion deep learning operations per second on top of initial 1.5 teraflops of power.
NVIDIA also stated that the Parker System could also encode and decode 4K video streams running at 60 frames per second (FPS) which, of course, is a tremendous accomplishment on its own.
However, if we compare the Parker system with NVIDIA’s other deep learning programs such as the DGX-1 then the Parker System starts to look a tiny bit less impressive.
The DGX-1 is NVIDIA’s deep learning initiative for OpenAI (founded by Elon Musk in December of 2015) and it is capable of hitting 170 teraflops of performance.
That’s about 150 times more powerful than the Parker System which NVIDIA revealed a couple of days ago.
The DGX-1 would exclusively be used for running high-end digital dashboard and keeping future self-driving autonomous cars working in top condition.
NVIDIA also announced that the company had added more partners to its deep learning programs.
NVIDIA initially signed up Volvo to put NVIDIA DRIVE PX2 into Volvo’s XC90 car but now, as the company has stated, the company has struck deals with 80 carmakers, tier 1 suppliers, and university research centers.
Of course, this is just the start and NVIDIA will need to prove that it can bring in more customers to adopt its DRIVE PX2 and the Parker System technology sooner rather than later.