Tesla Owner Successfully Upgrades HW3 Model 3 to HW4

Tired of waiting for official hardware retrofits for older vehicles, one well-known developer in the Tesla community has taken matters into his own hands. Michał Gapiński, the software engineer behind the Tesla Android project that unofficially brings Apple CarPlay to Tesla screens as Tesla continues to work on an official implementation, has successfully retrofitted a Hardware 4 (HW4/AI4) Autopilot computer into an older 2022 Model 3 originally equipped with Hardware 3 (HW3/AI3).

The project proves that upgrading older Tesla vehicles to the current self-driving hardware is mechanically viable without replacing entire wiring harnesses or major modifications. Gapiński, who previously also retrofitted the Model Y L’s faster, air-cooled wireless chargers into a regular Model Y, took to X to share his work. He noted that he successfully transplanted a HW4 computer into the older Model 3 and even got the first camera to work.

“It’s not as bad as it seemed,” said Gapiński, adding that the computer replacement only needed “some light ECU hacking.” According to the tinkerer, the camera upgrade is possible with adapters alone —there was no need to replace entire camera cables or harnesses. “It’s driving, but still a bit of stuff left to get all the features,” he continued, concluding that Tesla could already offer this retrofit as a first-party solution to owners since it doesn’t require any new parts other than the HW4 computer and cameras.

Inside the HW4 Computer Retrofit

Making a newer self-driving brain talk to an older vehicle required a mix of hardware ingenuity and firmware adjustments.

Gapiński sourced a HW4 computer originally pulled from a Model Y. Describing the mounting process, he joked that it only took epoxy, zip ties, and a YOLO attitude. He noted that because the casing is only slightly taller, an official “Highland” refresh bracket would also bolt up cleanly if you drill mounting holes.

@mikegapinski

A major technical question had been whether older camera cables could support newer HW4 sensors. Gapiński confirmed:

“It’s both Fakra, you just need to adapt it to a different connector. No need to cut and replace anything.”

While all external cameras must be swapped out for newer HW4 units, using custom adapters allowed the factory wiring to remain intact. Gapiński noted that cooling and power delivery in the 2022 Model 3 appear adequate to support the HW4 computer, though he added that upgrading older Intel Atom infotainment units to AMD Ryzen is actually easier than swapping the Full Self-Driving computer and camera suite.

Bridging the HW3 Performance Chasm

The motivation behind the swap comes down to raw compute power. Hardware 3 has only about 15% of the memory bandwidth found on Hardware 4. That bandwidth limitation left HW3 cars stuck on older FSD v12.6 builds for over a year before Tesla publicly released FSD v14 Lite last month.

However, while FSD v14 Lite is a major upgrade over v12.6, it still lags behind the main v14 branch in confidence and decision-making speed. What’s more, running modern neural networks on older silicon is pushing the hardware to its thermal limits, with reports of HW3 computers overheating on FSD v14 Lite.

Tesla Upgrade

The upgrade from HW3 to HW4 has been a big topic of discussion over the years with many suggesting it wasn’t possible or would require major modifications. However, what we’re seeing here is an owner able to complete the upgrade himself, without any support from Tesla.

Tesla has been relatively quiet on the HW3 upgrade, but they have said it is coming for owners who purchased FSD in full. Most recently, they confirmed that it will also require an upgrade to the higher-resolution HW4 cameras and that Tesla would likely have to open up temporary upgrade centers across the country.

They also said an upgrade doesn't make sense right now until Tesla can solve autonomy and knows the hardware required to achieve it.

Not a Tesla App

Tesla also recently confirmed that HW3 vehicles won’t achieve unsupervised autonomy. Yet, with no official retrofit program in place and Tesla facing a class-action claim in Europe from frustrated HW3 owners, community tinkerers are filling the gap.

As newer silicon revisions arrive, including the upgraded AI4+ (or AI4.5) FSD computer that Tesla teased earlier this year, DIY projects like Gapiński’s prove that older Tesla vehicles don’t have to be left behind by modern autonomy stacks.