Finally someone did it! Tesla Model S Autopilot 2.0 computing unit is completely disassembled, what does the Nvidia PX 2 motherboard look like?

Leifeng.com New Intelligent Driving Note: In October 2016, Tesla released the new "HW2" sensor suite and Autopilot 2.0 system based on NVIDIA's Drive PX2 computing platform, and installed the new computing unit NVIDIA PX2 AP2.0 in the panel above the car's glove box. Its disassembly and installation will not affect other functions of Tesla models. Some media have said that this is in line with Tesla's philosophy, that is, it does not rely on any platform and can be driven by any third-party processing unit.

However, whether Tesla abandons Mobileye now or develops its own computing platform in the future, we all know that Nvidia PX2 is Tesla's choice at this stage. So what does Tesla's visual brain look like now? From the release of the new kit to now, very few people have actually disassembled the Drive PX2 processor to find out. And just a while ago, a car owner from the Tesla Club Forum finally practiced this process. Leifeng.com New Intelligent Driving has organized the whole process of his disassembly of the PX2 AP2.0 computing unit as follows for readers.

Since the release of Tesla's Autopilot 2.0 system, no one has taken down the new Nvidia PX2 AP2.0 box to find out what's inside, but today, I did it. Previously, many people speculated that Tesla uses a dual-slot CPU and GPU motherboard, which is also the product feature that the PX2 platform has always advertised. Let's take a look first. Nvidia's official website lists two motherboards, one single-slot and one dual-slot, as shown below.

Many people assume that Tesla uses the dual-slot "Autochauffeur" version in the picture above because the motherboard has 8 camera connection points, which makes people think it is tailor-made for Tesla's HW2 hardware package. But in fact, the single-slot "Autocruise" motherboard in the picture above can also support data input from sensors such as 8 cameras (and millimeter-wave radar/lidar).

After disassembling my Tesla myself this time, I was able to completely confirm that Tesla was using the "Autocruise" motherboard, with 8-way camera data input and millimeter-wave radar input, and a single-slot CPU and GPU.

As you can see, the circuit board design is specially customized for Tesla, and the diffuser housing is larger than normal. I guess this is because this version is the first generation product developed by Tesla based on Nvidia's PX2 computing unit, and the larger diffuser housing is in preparation for the subsequent switch to a dual-slot motherboard. In other words, they have prepared a larger motherboard redundancy for AP2.0 during development, but what we have now is part of the functions of the small motherboard.

Now we start the disassembly. First, here is the exterior of the car, the Model S 2016.12 S60 RWD (rear-wheel drive) version.

First, disconnect the power supply through the car's main screen, then open the glove box in the car (that is, the storage box under the co-pilot instrument panel). If the power is not disconnected, the storage box is locked. Then, disconnect the 12V power supply from the negative terminal of the battery, and then disconnect the emergency power cord.

Remove the upper and lower sealing edges of the glove box.

Remove the six screws securing the glove box, remove the three wires connecting the glove box, and you can get the computing unit!

Now, you can see the AP2.0 computing unit exposed, which is the silver box. Disconnect the connecting wires around the box and unscrew the four fixing screws. Then you can completely remove the AP2.0 processing unit.

So, what will you find when you open this box?

This is the complete PX2 motherboard.

The above is the complete process of disassembling the HW2 AP2.0 system NVIDIA PX2. The version number of this device is 1078321-60-C.

Recommended reading from Leifeng.com: In-depth analysis: How is Tesla Autopilot 2.0 better than Autopilot 1.0?