Announced the Nvidia Drive PX Pegasus for fully Autonomous cars
For the second year technology conference GPU Technology Conference (GTC) held not only in California, where she traditionally held a few years each spring. Starting in 2016, NVIDIA has expanded its presence in the world of software developers, conducting a conference, not only in Silicon valley but also in China, Germany, Israel, Taiwan, Tokyo and Washington.
Today, GTC Europe started its work in Munich, and it is not surprising that in the car the edge of Europe, the lion’s share of performances and announcements devoted to the subject of cars. In particular Autonomous vehicles, in the development and promotion of which NVIDIA is directly involved for several years. They even have their own Autonomous vehicle BB8 designed for running. To date the best product for avtopilotiruemyh car was the NVIDIA DRIVE PX 2.
Today in Munich Jensen has announced a new product that became the world’s first hardware-based solution suitable for creating a fully Autonomous robotic taxi. The new platform was named Pegasus, and it extends the capabilities of computing platforms, the NVIDIA DRIVE PX to the fifth level of autonomy fully Autonomous vehicles, requiring no human intervention to manage. The NVIDIA DRIVE PX Pegasus delivers performance at 320 trillion operations per second, more than 10 times faster than its predecessor — the NVIDIA DRIVE PX 2.
Platform Pegasus makes it possible to create a new class of vehicles running without a driver and without controls: steering, pedals, etc. Such robotaxi will arrive at the passenger’s request and deliver it to the destination. Such cars will allow you to optimize traffic flows, Parking spaces and will reduce accidents on the roads.
Platform DRIVE PX has received broad support from industry, almost all manufacturers and related companies use solutions NVIDIA in the development of vehicles 5 level of autonomy. Out of 225 partners working with the NVIDIA DRIVE PX, more than 25 companies are developing a fully Autonomous, robotaxi, using the capabilities of the desktop GPU NVIDIA. Racks such working prototypes downloaded all server racks on the basis of several NVIDIA graphics processors that exploit the capabilities of deep learning and other algorithms of parallel computing. Of course, this solution is not suitable for the production car, but only for testing of technologies in practice.
Compute abilities of Autonomous platforms is just huge, because the processing and analysis of data with multiple cameras and lidars, and high resolution, and tracking of objects in space with high precision requires a large number of calculations. And they must all be performed accurately, because it affects the safety of others. If you compare the modern cars with the beginnings of the capabilities of the autopilot with fully unmanned robotaxi, the latter requires 50-100 times more calculations.
And these possibilities are provided by Pegasus DRIVE PX, which is based on four high-performance chips NVIDIA: two new systems-on-Xavier-Chir, the graphics core which is based on the architecture Volta and two discrete GPU of the next generation of specially optimized for machine vision applications and deep learning. It is not surprising that the platform has the processing performance 320 TOPS (of operations used in deep learning), and enables the creation of fully Autonomous, robotaxi on the basis of the specialized system of small size having a relatively low power consumption of 500 watts. But the combined memory bandwidth exceeding 1 TB/s.
By the way, it is possible to calculate the approximate performance of the GPU of the next generation. If the total calculation speed is 320 TOPS, 60 of them have a couple Xavier and the remaining 260 are distributed between the two GPUs. It turns out that the performance of one GPU is equivalent to 130 TOPS — slightly higher than in Volta V100 with performance 112-125 TOPS. But the power consumption of one system is 250-300 watts, and entire Pegasus, consisting of a pair of Xavier and a pair of unnamed GPU, satisfied with 500 watts.
The Pegasus platform is designed with ASIL D certification — the industrial security standard, and support CAN-bus, Flexray, 16 high-speed inputs for cameras, radar, lidar and other sensors, and multiple 10 Gigabit Ethernet connections. Platform DRIVE PX does not have to consist of four processors, it massturbate from one system-on-chip that provides opportunities for levels 2 and 3 of autonomy to full chetyrehjadernogo decision of the SoC and GPU that supports level 5. All configurations use the same software architecture and that gives the possibility of simultaneous development of various products — from the possibility of using a partial auto-pilot on highways to Autonomous robotics.
Although the Pegasus platform is based on a yet to be announced GPU is still unknown to the public architecture, NVIDIA announces availability of the platform for partners in the second half of 2018, and early access will be at the end of the first quarter of next year. On the one hand, up to this point for a long time, on the other — yet on the horizon I can not see any competitor for the Pegasus, even with this in mind. Moreover, the development of can lead now using the other NVIDIA solutions, because the software DriveWorks and early decisions DRIVE PX 2 is available for developers today.