Twitter user @greentheonly disclosed pictures of Tesla’s 4th generation Autopilot hardware HW 4.0. The size of the 4.0 hardware has increased and the interface is completely different compared to HW 3.0. Musk had previously said that HW 3.0 models do not support the upgrade to 4.0.
Regarding the specifications, HW 4.0 is still based on Samsung Exynos IP, with 20 CPU cores compared to the previous 12, a maximum frequency of 2.35 GHz, and an idle frequency of 1.37 GHz. The TRIP deep neural network accelerator has been upgraded from dual-core to tri-core, with a maximum frequency increase from 2.0 GHz to 2.2 GHz. The estimated total computing power is between 300-500 Tops. Tesla has also increased the number of camera interfaces from 9 to 12, including 1 backup, 7 Autopilot (2 front, 2 left, 2 right, 1 rear, 1 front looking, 2 rear looking, 1 inside the car). In addition, Tesla has added a new GPS module with 3-frequency antennae, radar interfaces and Phoenix millimeter-wave radar heaters, with radar communication being 100 Mbps Ethernet.
The redundancy design of HW 4.0 hardware is relatively complete. The motherboard adopts a symmetrical design, with two identical power interfaces, and the information entertainment (central control) also has two network interfaces. The information entertainment motherboard has been redesigned, with GPU and CPU integrated together. Power consumption has doubled, idle power consumption is up to 80 W, ensuring that the vehicle can still operate normally in special situations, while using liquid cooling for heat dissipation. In addition, there should be some integrated designs, as it can be seen that 4.0 hardware is thinner than 3.0. We expect that the detailed information and parameters of this hardware will officially be announced at Tesla’s press conference on March 1.
This article is a translation by ChatGPT of a Chinese report from 42HOW. If you have any questions about it, please email bd@42how.com.