NVIDIA Unveiled New DGX SuperPod


Jun. 30 2019, Updated 2:48 p.m. ET

Autonomous vehicles create the need for DGX SuperPod

NVIDIA (NVDA) unveiled its new DGX SuperPod supercomputer for simulation. The company came up with the idea of developing such a supercomputer as it was faced with the computational problem of training autonomous cars.

One big issue with autonomous vehicles (or AV) is to train them. It is difficult and time-consuming to make an AV complete a certain amount of driving hours on different types of roads. There are also physical limitations in training the AV on different real-life scenarios like extreme weather conditions, human error, kids running out from between cars, flat tires, or driving alongside an unsafe driver on the road. Moreover, AV has to be trained at the speed a car drives, regardless of how fast the computer inside the car is.

The end goal of building autonomous cars is to reduce accidents. Thus, there is little tolerance for accidents by autonomous cars. The physical limitations of training cars on roads and the high safety standards created a computing need for a virtually simulated training environment.

Article continues below advertisement

Simulation: the solution to autonomous vehicles training

To address the issue of training AVs, NVIDIA created its DRIVE Constellation platform, which allows developers to create different scenarios in a virtually simulated environment and accelerate the training process of AV. Moreover, developers can train the AV at the computing speed of the system inside the vehicle. The DRIVE Constellation platform solved the issue of training the AV for roads. However, NVIDIA faced another issue of continuously retraining the AV.

In order to drive safely, an AV needs to factor in all of its surroundings instead of focusing on just one thing, which amounts to emulating one terabyte of data generated from its video, LIDAR, radar and sensors every hour. Apart from emulating its own surroundings’ data, AV has to retrain itself continuously over time using the data from an entire fleet, which could go up to petabytes (1000 terabytes=one petabyte).

Thus, NVIDIA developed DGX SuperPod to meet the computing need for retraining. At present, the AV market is still in the development stage and are exploring its needs. If NVIDIA felt the need for a DGX SuperPod computer, other automakers developing AVs likely will too, as they would like to accelerate the process of putting AV on roads.


More From Market Realist