Tesla Dojo fabric lead Eric Quinnell has left the company.

Quinnell was behind the Tesla Transport Protocol Over Ethernet (TTPoE), a new lossy, exascale fabric developed for Tesla's own custom Dojo supercomputer effort.

TTPoE is a peer-to-peer ethernet transport layer protocol that is executed entirely in hardware, which the company says has a much lower latency than TCP/IP. The Dojo supercomputer features custom Dojo Training Tiles, and Tesla's Mojo 100Gbps NIC.

Quinnell spent nearly four years at Tesla, after stints at Arm, Samsung Austin R&D, and AMD. At AMD, he worked on the low-power x86 architecture that went into the PS4 and Xbox One.

He is set to join Amazon Web Services to work on the company's Trainium AI chip.

He joins a number of departures from Dojo, following the head of the effort leaving last December. Soon after, the head of AI infrastructure joined Google DeepMind. This June, Reza Khiabani, the head of Dojo's cooling infrastructure left to join OpenAI.

Others were laid off as part of wider cuts earlier this year, while some Tesla AI researchers have moved to Elon Musk's private xAI venture.

However, a significant number of employees remain at Dojo, and the project is still very active, led by Peter Bannon. Alongside the custom chip initiative, Tesla is procuring a large number of Nvidia GPUs to help with autonomous vehicle training.

In July, Musk said that the company was "going to double down on Dojo, and we do see a path to being competitive with Nvidia with Dojo," at the same time as building out a 50,000 H100 GPU cluster.