Tesla starts production of Dojo supercomputer to train driverless cars
1–7 of 7 posts
Re: Tesla starts production of Dojo supercomputer to train driverless cars
#2Ah, the old "throwing more hardware at a problem will surely solve it" approach.
Re: Tesla starts production of Dojo supercomputer to train driverless cars
#3Ah, the old "throwing more hardware at a problem will surely solve it" approach.
Nah, this is just run of the mill stock pumping in preparation for when Tesla acquires xAI and/or Twitter for their invaluable AI technology that is totally the key to FSD.
Re: Tesla starts production of Dojo supercomputer to train driverless cars
#4Ah, the old "throwing more hardware at a problem will surely solve it" approach.
Indeed. The common answer for scaling compute is the recruit more hardware.
Tesla has a vast amount of video data to feed into its vision --> action model. Training on Nvidia hardware takes too long. Dojo fixes this problem by a novel SIMD architecture.
Re: Tesla starts production of Dojo supercomputer to train driverless cars
#5Ah, the old "throwing more hardware at a problem will surely solve it" approach.
Indeed. The common answer for scaling compute is the recruit more hardware. Tesla has a vast amount of video data to feed into its vision --> action model. Training on Nvidia hardware takes too long. Dojo fixes this problem by a novel SIMD architecture.
The whole concept of a GPU is performing SIMD, what's novel more precisely?
Re: Tesla starts production of Dojo supercomputer to train driverless cars
#6Earlier quoted context omitted.
Indeed. The common answer for scaling compute is the recruit more hardware. Tesla has a vast amount of video data to feed into its vision --> action model. Training on Nvidia hardware takes too long. Dojo fixes this problem by a novel SIMD architecture.
The whole concept of a GPU is performing SIMD, what's novel more precisely?
He’s just spouting Tesla propaganda. Check out his comment history.
Re: Tesla starts production of Dojo supercomputer to train driverless cars
#7Ah, the old "throwing more hardware at a problem will surely solve it" approach.
Well, yeah. Deep learning shows that approach works (as long as you have the data as well).