Nvidia wants its new Drive PX 2 supercomputer to drive your next car
Nvidia wants its new Bulldoze PX 2 supercomputer to drive your next car
Until now, most cocky-driving cars take required an array of expensive, custom, processing hardware. Nvidia is aiming to change all that with the introduction of its Bulldoze PX two liquid-cooled, deep-learning, monstrously powerful computer that tin can easily fit in the torso of your next car. Nvidia is using CES 2016 to deliver on the vision it teased last year: that cars would be more almost software, that cocky-driving will require deep learning, and that all the systems required will need to run in real fourth dimension. Its new Drive PX ii is designed to evangelize the goods — information technology features the power of six Titan X boards when used for the type of sophisticated processing needed for deep learning and autonomous vehicle operation. Along with the Bulldoze PX 2, Nvidia's co-founder and CEO Jen-Hsun Huang rolled out an extensive platform for autonomous vehicle organisation development using Nvidia's tools.
Nvidia's democratic vehicle platform
Like almost all electric current self-driving car architectures, Nvidia's starts with massive cloud-based deep-learning compute ability. To hear Huang tell it, everyone has given upwards on traditional object recognition algorithms and is moving to the booming technology of deep learning. Deep learning is characterized past software that determines its own features for employ in solving recognition and other problems. Previously, for tasks such as object, gesture, and facial recognition, developers had to painstakingly identify characteristics that could be measured and used as features to assist their software acquire. Improving results meant hand-tuning those features. When processor power was more than limited, that was the only practical approach.
Now that inexpensive GPUs have made parallel processing amazingly cheap, systems where the software can identify what features to apply — deep learning — have go not only practical, just seemingly inevitable. Modern GPUs can learn at xxx-40 times the rate of previously available platforms. Suddenly deep neural networks that can recognize objects fifty-fifty faster and with more accurateness than humans — envisioned equally early equally the 1980'due south — are practical.
Nvidia calls its system for automotive deep learning DIGITS. It's designed to pipe its results to the runtime DriveWorks platform powered by its Bulldoze PX ii. In that location is one sleight-of-manus hither, though. Developers nevertheless needs to decide which inputs to mensurate, and how to calibrate and quantify them before feeding them into the software. It's not like a HAL 9000 that tin be unleashed on the globe without guidance and suddenly learn how to drive or sing.
Whatever solution automakers develop, Nvidia wants to brand sure it volition run on its processors. It has worked hard to ensure that essentially every major deep learning software toolkit can run using its CUDA platform. In plough, CUDA is binary uniform on Nvidia processors ranging from Jetson to Titan Ten, and at present Drive PX. Whether information technology'southward the usually-deject-based training software– Nvidia DIGITS — or the vehicle runtime — Nvidia DriveWorks — Nvidia wants to ensure that its processors are involved.
Even with plenty of training washed in the cloud, processing requirements for the vehicle itself are substantial. Real time object recognition, image fusion to create a 3D model of the surroundings, integration of map data, and responding to the deportment of nearby vehicles all require plenty of compute horsepower. That'south why Nvidia has connected to push the performance envelope of the Drive PX with the much-more-powerful Bulldoze PX 2.
Drive PX two Unveiled
Specifications for the Nvidia Drive PX 2 are listen-blowing. It incorporates roughly the horsepower of 150 MacBook Pros when information technology comes to compute-intensive issues. With viii Teraflops of floating point ability, and 24 TOPS (a mensurate of deep learning power that compares to a half-dozen Titans), the Drive PX 2 tin manage a benchmark of identifying ii,400 images per 2d running the award-winning Alexnet software (compared to 450 per 2d with a Titan X). Huang describes the result as $300,000 worth of computers in a lunchbox. The Drive PX 2 is liquid-cooled to allow it to run in the harsh environments to which vehicles are often exposed. For cars that don't accept liquid cooling an optional Heat Transfer Unit (basically a fan) is available every bit an option.
A vision of cocky-driving without the driving
Conspicuously absent-minded from Nvidia's vision of a self-driving automobile is actual driving. Clearly Nvidia views its strength as the collection and analysis of the data needed for a driving system to brand decisions on dispatch, steering, and braking. Nvidia has already announced that Volvo will be its first partner for integrating the Drive PX ii into 100 XC90s every bit function of its autonomous-vehicle program. But Nvidia — probably wisely — stops curt of addressing the issue of vehicle command. On the ane hand, this is a clever strategy, every bit it makes it easier for Nvidia to focus on its strengths and to partner with more than traditional car companies — and even with new players like Apple and Google. On the other paw, it leads to a false sense of optimism when slick demos of cars gliding through traffic aren't fleshed out with actual control systems to make them a reality. The skilful news for the autonomous vehicle marketplace is that at that place is now a supercomputer optimized for making self-driving a reality that can comfortably fit in your trunk.
Now read: Artificial neural networks are changing the world. What are they?
Source: https://www.extremetech.com/extreme/220339-nvidia-wants-its-new-supercomputer-to-drive-your-next-car
Posted by: christensenwele1946.blogspot.com
0 Response to "Nvidia wants its new Drive PX 2 supercomputer to drive your next car"
Post a Comment