What is tesla’s “dojo”?


Tesla’s “Dojo” is the name of the learning supercomputer of neural networks, which is an AI algorithm. Because it is a place to train, it seems to be named “Dojo” = “Dojo”.

Dojo itself is still under development, but tesla’s head of AI announced the prototype supercomputer.

“this is a really great supercomputer”

🟩 For AI for autonomous driving

Tesla will use its dojo supercomputer to learn its self-driving neural net. Tesla owns about 10 seconds to 1 million scenes and data tagged with 6 billion objects. With 1.5 petabytes of data, it can be seen that learning autonomous AI requires a huge data set.

It is inevitable for Tesla, which aims for an autonomous driving system that relies only on cameras with large amounts of data, rather than radar and rider sensors. Therefore, a supercomputer that processes a huge amount of data is required. In fact, Tesla has already changed its Model 3 and Model Y for North America to be non-radar, increasing the dependence of cameras.

🟩 What is the performance of the supercomputer “Dojo”?

The unveiled Dojo prototype uses a 20-node 8x NVIDIA A100 Tensor core GPU (5,760 GPUs in total). The performance is 1.8 Exa FLOPS, which is the fifth supercomputer in the world. However, Tesla has not yet run the benchmarks needed to enter the TOP 500s rankings.

🟩 There are plans to open “Dojo” outside the company

In April 2021, Tesla CEO Elon Musk indicated that he would open “Dojo” to external companies in the future.

🟩 summary

Tesla’s ‘Dojo’ is a supercomputer for the development of autonomous AI

By using “Dojo” by external companies, Tesla will accumulate know-how and the algorithm of autonomous driving will be refined.

Dojo currently uses NVIDIA’s GPU, but in the future, Tesla’s own Chip may be an option.


Thank you for watching. Please share if you like!