No, the "self driving computer" is independent and it actually runs itself in parallel for redundancy and error checking.

The infotainment system runs a custom linux distribution running on ARM (MCU1) or intel atom (MCU2).

They talk to each other but they are independent.

Question is, how do they talk? For example, if they have shared memory on same bus, it is BAAAAAD.

They talk over the CAN afaik.

It's false. The infotainment computer (MCU) is separate from the autopilot computer. You can even reboot the MCU while driving on autopilot and autopilot will keep chugging along happily--though I very much don't recommend doing this since you lose all your feedback from the car (speedometer, audio, ...) while it's rebooting.

There are two versions of the MCU and 3 or 4 versions, depending on how you count, of the autopilot computer.

They are wrong. The ECU driving the car is separate from the computer running neural net inference (either the NVIDIA drive PX or their newer FSD computer), which is in turn separate from the display or "MCU" which controls the entertainment system which is based on an Intel Atom CPU.

Auto pilot still works when you restart the infotainment and display system (holding two buttons on the steering wheel down), so the systems aren’t powered by the same OS.

It is only half of the story. I was almost sure, that ECU and Infotaiment runs on different CPUs under different OSes. Question is, what do they share? Same bus? Some memory area? Or only something like I2C?

I could imagine situation, when hacked infotaiment system overloads power supply, and if power supply is shared, it takes down ECU too, for example.

I hope, Tesla's engineers better than that.

> that ECU and Infotaiment runs on different CPUs under different OSes. Question is, what do they share? Same bus? Some memory area? Or only something like I2C?

That's not how auto engineering works. Read up on Controller Area Network (CAN)Bus. Tesla MCUs (Touchscreen) are on a completely separate sub-system from the power-train/braking systems.

> I hope, Tesla's engineers better than that. Sure. Most of Tesla engineers came from other auto manufacturers. Lot's of them were form the UK during the development of the Model S.

This is absolutely false. The new FSD chip is specifically made for Neural net inference tasks. It has a custom instruction set with ~18 instructions. Even before that, they used an NVIDIA Drive PX unit for NN inference.

The in-car entertainment system is run on a separate Intel Atom based computer. The ECU controlling the car itself is completely separate from both of these. You can reboot the main display of the car while driving on Autopilot and it will still perform the same.