NVIDIA AI Hardware Launch Cements Autonomous Driving Leadership

Already a leader in mindshare around artificial intelligence and self-driving technology, NVIDIA announced at its GPU Technology Conference a new platform it promises will power fully autonomous vehicles. The Drive PX Pegasus is a high performance system that combines ARM-based processing cores with two yet-to-be-announced graphics chips offering as much as 10x better performance than the previous best offering from the company. Though sampling and shipping won’t start until the second half of 2018, NVIDIA has already announced a partnership with Deutsche Post DHL and ZF to bring delivery trucks with full autonomy to roads in 2019.

As with previous Drive PX systems, NVIDIA is targeting the dozens of automakers and auxiliary technology companies that are working to build and create self-driving cars and infrastructure. This includes current customers like Tesla, Toyota, Audi, Volvo, and a host of secondary adopters like Uber, Lyft, and Google. Though the holy grail of this technology will see a self-driving car in every garage, NVIDIA claims it will start with roll-outs in smaller, contained areas like college or business campuses, airports, and even large leisure spots like Walt Disney World. As the acceptance of self-driving vehicles grows, the leaders like NVIDIA will be able to address enormous markets including taxis, emergency response, and shipping, presenting a gold mine of opportunity.


Drive PX Pegasus clearly defines NVIDIA as the leader in the autonomous driving space from a technology and capability standpoint. It has dominated in compute performance for the gaming, graphics, enterprise, and AI fields for a several years, and translating that to self-driving cars is a move that will help steer the direction of the GPU giant going forward. There is still risk from companies that may decide to vertically integrate, like Google and Tesla, but NVIDIA has created a significant lead over its competition with partnerships, development, and the first true integrations in consumer products. Though Intel has used acquisition to speed up its own development in the field of autonomous driving, it has a colossal task ahead to catch up to what NVIDIA CEO Jen-Hsun Huang has been to achieve through adaption of its graphics processor portfolio.

With a target ship date of 2019, the NVIDIA announcement does bring into question that validity of Tesla’s previous claims that it would be able to offer full Level 5 vehicle autonomy (the highest rated level requiring no driver invention, as stated by SAE, a standards body for the automotive industry) with current shipping cars and hardware. With a 10x improvement in performance on the new NVIDIA Drive PX Pegasus hardware, I expect safety advocates to begin questioning the power and capability of the systems in the Model S and Model X, both of which are currently powered by older NVIDIA hardware.

This debate may also be a reason for the move by Tesla to bring chip and processing development in house, as rumored last month.


From a technological perspective, Drive PX Pegasus is clearly a forward-looking device. Rated at more than ten-times the performance of the Drive PX Xavier NVIDIA announced earlier this year, it is capable of more than 320 trillion operations per second, an important trait when working with a stream of data from the system’s supported 16 cameras. Two custom ARM-based Xavier chips and two unannounced architecture-based discrete (separate) graphics chips provide the processing power to decipher the imagery input, manage driving tasks, and monitor vehicle surroundings.

NVIDIA has only just started shipping graphics processors based on the company’s Volta architecture for AI workloads and enterprise applications, but Drive PX Pegasus will use a design even newer than that. This indicates that there will be at a long wait before seeing mass production of the new autonomous vehicle hardware, with extremely low volume sampling as much as a full year off. Whether or not this gives competing solutions enough time to incubate has yet to be seen.