Can anyone tell if this architecture makes sense to ship in their actual cars to replace whatever is doing the computation there now? That is the only way I can see this making any sense to develop.
They have already developed and deployed custom silicon for their cars.
There is probably some overlap in the IP between these chips, but Dojo is optimized for operating in a large data center cluster for training purposes, whereas their car chips are optimized for running pre-trained models.
This is a chip designed to accelerate nn training in their datacenter.
The effort and expense makes perfect sense if you consider:
- the costs are spread over millions of car (3 millions, today, tens of millions in the future)
- it helps them win robotaxi market that is potentially so lucrative that the cost of developing custom chips will be peanuts