In a weekend speech at the International Conference on Computer Vision and Pattern Recognition (CVPR), Andrej Karpathy, the head of Tesla's artificial intelligence team, demonstrated a supercomputer .
This supercomputer uses 5760 Nvidia A100 graphics cards with a computing power of 321TFLOPS to form 720 nodes, with a total computing power exceeding 1.8EFLOPS and 10PB of storage space.
It is said that Tesla's supercomputer will be used to process the autonomous driving data of its vehicles .
In the field of autonomous driving, Tesla has always bucked the trend.
When Waymo or Weilai and other teams dedicated to autonomous driving are vigorously developing lidar to achieve autonomous driving, Tesla insists on using cameras and focusing on visual recognition .
Tesla collects driver and vehicle data through various sensors in the car and uploads it to this supercomputer, so that its artificial intelligence software can analyze the data and learn how to drive like a human.
There are countless Teslas on the road today. To process massive amounts of data from all Teslas, powerful computing power is required. The supercomputer shown by Andre is currently in use. It is said that its computing power is close to Selene, the fifth largest supercomputer in the world.
Andre admitted in the interview that vision-based autonomous driving will be more difficult to achieve than autonomous driving with lidar, which uses lasers to identify obstacles around the car. The camera used by Tesla will face certain risks in some cases, such as foggy weather.
During Tesla's early fully automated driving tests, the system often made misjudgments and required manual intervention by the driver.
However, Andre believes that adding lidar to Tesla may not be a good choice:
Using lidar and camera at the same time will cause a problem for the software, that is, when the two differ in what they see, it may be difficult for the system to determine which sensor is correct.
If you focus on traditional cameras, Tesla can improve computer vision to train the computer how to deal with the unpredictable behavior of pedestrians and passing vehicles, so there is no need to use expensive laser sensors.
Interestingly, it was discovered that Tesla was conducting a lidar test in Palm Beach, Florida, USA. In the photo, you can see a Tesla Model Y equipped with lidar.
It is reported that Tesla is using Liminar lidar to verify the reliability of its camera-based fully autonomous driving.
▲Picture from: Grayson Brulte
At present, Tesla has millions of electric vehicles on the road, providing useful data for them, and its supercomputer can analyze and train based on these data.
However, Tesla's Autopilot automatic driving assistance system is far from reaching the stage of "automatic driving", so it remains to be seen whether its plan is correct.
The title picture comes from: Motor1
#Welcome to follow Aifaner's official WeChat account: Aifaner (WeChat ID: ifanr), more exciting content will be provided to you as soon as possible.