Tesla constantly evolve the MCU and Autopilot (AP) hardware used in their cars. This guide describes the technical differences between the key versions of Tesla hardware and supports our guide to the Functional differences between Tesla hardware versions
Tesla have run two broad versions of autopilot hardware, the original Mobileye version, sometimes known as AP1, and the subsequent Tesla version which was originally called Enhanced Autopilot (EAP), and latterly reduced to just Autopilot (AP) and Full Self Driving (FSD).
The first cars had no Autopilot hardware and on the early cars even parking sensors were an option later becoming standard as part of the technology pack. It was only when the Technology pack with convenience features came out that AP hardware started being included although at the time Autopilot was still some way off.
AP1 hardware was first introduced as part of the technology pack with convenience features and later became standard fit regardless with the introduction of the dual motor cars such as the 85D. Tesla bought in AP1 technology from Mobile eye and is based on their EyeQ3 hardware which is also used by a number of other makes of car for active cruise control, lane departure warning etc. Tesla pushed the capability of the EyeQ3 hardware further than anyone else and in part that was why Mobileye and Tesla fell out. Tesla essentially push technology to breaking point and that results in failures, Joshua Brown being the first fatality as a result of relying on the autopilot system beyond its capabilities. While Tesla do stress the versions are driver aids, they also promote full self-driving and do little to ensure drivers use the systems correctly. This has set a pattern of behaviour of Tesla ever since, pushing beyond the capabilities of the hardware and software to give a 99.9% positive experience and hope the 0.1% of poor experience is not catastrophic.
AP1 hardware includes
Mobileye provide a turnkey solution with much of the processing done within the camera. As part of the solution certain features such as speed limit reading were included as standard, something that Tesla are yet to introduce, partly thought to be because of patent issues.
Tesla released their first system at the end of 2016. While Tesla are unlikely to ever admit this, the introduction of the hardware was premature and done as a reaction to Mobileye contract being terminated. Tesla promised that the hardware was capable of Full Self Driving, they included a demo of a car driving from home to work autonomously and claimed this would be possible "soon".
HW2 hardware includes
Tesla adopted a different approach to Mobileye by centralising the processing with each camera sending a video feed to the central processing unit.
The second generation of hardware including a number of detail changes including:
Tesla claim the changes are more tweaks and preferred to say it was more like HW2.1 than HW3. Tesla also still maintained the hardware was all you needed for full self-driving and HW2 owners would get a free upgrade if they had purchased the FSD package.
The third generation of hardware was more substantial and has brought in much more processing capacity with a switch to Tesla's own custom hardware.
The introduction of HW3 has certainly resulted in a number of extra features being introduced. Talk of processing capacity is only part of the story as code is optimised for a given processor and initially the autopilot software was optimised for HW2.5. There seems to be evidence that Tesla now produce different code for HW3 as it appears to be able to process things more quickly and resolve objects such as cones that AP 2.x cannot.
Tesla still maintain that HW3 will be capable of FSD and owners of the previous versions will be upgraded when the FSD features require it. This is becoming slightly semantic as we already see HW3 capabilities diverging from that possible on earlier versions but Tesla seem to adopting the policy that this is feature execution and not feature capability - ie a HW2 car can still steer down the road even if it would crash into a cone that a HW3 car would see and take into account.
Tesla have trialled upgrades to HW3 for some HW2.5 cars in the US but this is not widespread and its been reported that the change to MCU hardware during the life of HW2.5 has caused a number of challenges and MCU1 cars are currently not able to upgrade to HW3.
These are the approximate dates when the carious autopilot hardware was introduced. The difference versions have different configuration option codes except on the Model 3 which is still reporting HW2.5 in their configuration.
Tesla have had a number of MCUs over the years. With the exception of very early screens, there is MCU1 used up until around March 2018 and MCU2 which has been in production since then on their MS and the MX. The MCU is totally separate to the autopilot hardware and the performance of one is not dependant on the other.
This is the main screen in the centre of the S/X dashboard and includes the touch display, processors, memory, audio, WiFi, Bluetooth, GPS, CAN bus communications, etc.
The mobile phone module is an additional board which changed in 2015 from 3G to 4G with the 4G board a retro fit option.
Version 1 was an Nvidia quad core Tegra 3 (arm) processor running an Ubuntu distribution of Linux.
This does largely the same as MCU1 but uses an Intel multi core Atom E8000 series CPU (x86_64) processor. It also includes a number of updates including a later version of bluetooth, faster overall graphics and a faster Chromium based browser.
The architecture is different as the display and the processing part are separated however the Atom processor used on the MCU2 set up is also used for the M3.