Chestnut connects an external graphics processing unit to Comma.ai’s in-vehicle computer, increasing the available computing power for larger driving models while using open-source firmware.
Comma.ai has launched Chestnut, an eGPU dock designed to connect an external graphics processing unit to its in-vehicle computer. The dock addresses the growing computing requirements of larger AI models used by the company’s open-source openpilot driving system. Earlier systems operated within a roughly 10-W power limit, while Chestnut raises the available power budget to about 100 W.
Chestnut connects desktop GPUs to in-vehicle computers, allowing larger driving models to run inside the vehicle. The GPU can be mounted in the passenger footwell or beneath the passenger seat and powered from the vehicle’s 12-V supply. According to Comma.ai, the Chestnut class is intended to support the development of larger driving models in vehicles.
The first driving model introduced in the Chestnut class is openpilot 0.11.2. Compared with previous versions, it has 30 times more parameters and requires 100 times more floating-point operations per second. Comma.ai anticipates that Chestnut combined with its in-vehicle computer could achieve computing performance comparable to Tesla’s HW4 platform.
Another feature of the Chestnut hardware is its open-source firmware. The firmware is written in C and was developed by modifying firmware for the ASM2464PD controller, with the work available through the tinygrad GitHub repository. Making the firmware available gives developers an opportunity to examine and modify the software used by the Chestnut hardware.
Chestnut is available in two variants: the Tiny Chestnut Dock, priced at $249, and a version equipped with an 8-GB AMD Radeon RX 9060 graphics card, priced at $799. The Chestnut hardware can also be connected to a general-purpose computer, in addition to working with Comma.ai’s autonomous driving platform.
