Drive AGX Orin: Nvidia Accelerates on Autonomous Vehicles with “Next Gen” GPU Chip

[ad_1]

Nvidia has unveiled its new Drive AGX Orin chip for autonomous vehicles and robotics. It is capable of performing 200 trillion operations (10 power 12) per second, far more than the Tesla FSD SoC.

Known for its graphics processors, Nvidia is also very active in the autonomous vehicle with its dedicated platforms Drive PX, PX2 and Pegasus. On the latter rested two Xavier SoCs supported by dedicated GPUs for computing power that could climb up to 320 billion operations per second (TOPS in INT8)

Barely larger than an American license plate, the Drive PX Pegasus was designed for level 5 autonomous vehicles and more specifically for robot taxis, according to Jensen Huang, the very charismatic CEO and founder of Nvidia.

Recommended item: CES 2018 – Nvidia fully on the autonomous car (Xavier, Uber, VW …)

Orin: a chip of 17 billion transistors

Change of gear with a new platform called Drive AGX Orin which Nvidia does not hesitate to qualify as “Most advanced processor in the world”, capable of processing 200 TOPS on a single chip. What does the Orin chip contain? No less than 12 CPU cores and a “new generation” graphics circuit – of which we will know nothing for the moment. For the processor part, Nvidia gives up using its own internally produced cores and resorts to Hercules ARM cores. An interesting fact, because if Nvidia still refuses to speak of fineness of engraving, we can note that ARM intends these cores to engrave in 7 or 5 nm. Nvidia is currently capping at 12 nm on its most advanced chips, but we can estimate that the firm will be able to use 7 or 5 nm for its Orin platform, which is scheduled to be marketed for 2022.

This story of finesse of engraving takes all the more sense when we learn that the chip contains no less than 17 billion transistors, more than double the previous Nvidia chip intended for the automotive sector. We are thus not far from the 18.6 billion transistors of a Turing GPU as found on the GeForce RTX 2080 Ti, for example, and whose power consumption is close to 300 W.

NVIDIA is however confident, arguing that if Orin will be seven times faster to process data than a Xavier chip, it will also be three times more efficient in the energy field. Taking the efficiency data from the Xavier platform, we can expect an electrical consumption of around 70 W. Remains to be seen if Nvidia will undertake to build multi-chip platforms so as to further expand the computing capacity, as he currently offers it through his Pegasus platform.

Like its predecessor, Orin promises to simultaneously manage applications, but also artificial intelligence, the two bases that make up autonomous vehicles and robots. Available for car manufacturers from 2022, the Orin Drive AGX is intended for both level 2 and level 5 autonomous vehicles. Like Xavier, Orin will be programmable via APIs such as the recognition of traffic lights, pedestrians, trajectory perception and gaze detection.

This is going to relaunch the fight – friendly – that are engaged Tesla and Nvidia on SoC. Previously, the California automaker used Nvidia's chips, but is now developing its own SoCs, the latter of which is capable of performing up to 144 TOPS. It remains to be seen whether Nvidia's partners will be able to take advantage of the power of the Orin chip to stand out from Tesla systems.

[ad_2]