The future belongs to those who believe in the beauty of their dreams. -Eleanor Roosevelt
There's been a lot of talk by Tesla and Elon Musk for a while now about a new supercomputer, and on Tesla's Open AI day, the company introduced a model of its d1 chip tile.
So in this article, we are going to talk about the dojo supercomputer, what it is, why it was created, and how it works.
You can also check out
Before we get to the cool stuff, we are going to talk about the meaning of supercomputers and Tesla so we can get a sense of completion😉(emoji is pointless)
In simple words, a supercomputer is a computer that has a lot more processing power than your laptop at home.
Supercomputers are built for relatively large-scale tasks like flying a rocket, quantum modeling, predictive modeling, large-scale game theory, and other stuff.
In essence, a supercomputer is a computer built for complex, time-consuming tasks which can't be done efficiently on a normal computer.
If you are in the 21st century, you probably know what Tesla is.
In simple everyday words, Tesla is a company that makes electric cars, electric car batteries, solar panels, and everything in-between. That probably aligns with what you know Tesla as.
But according to Elon, Tesla doesn't see itself as a car company because of the unique way its cars are built. To paraphrase Elon, Tesla is actually a robotics company because to them; their cars are actually "semi-sentient" robots on wheels.
Like I said in
Now to the fun part
The Dojo is a supercomputer built by tesla to train its AI deep neural networks and Machne Learning algorithms.
In other words, Dojo is a supercomputer built by tesla to train its neural networks algorithms and Machine Learning models in a faster and more efficient way.
According to Elon, the Dojo supercomputer is going to be the fastest AI training supercomputer built. Why? We'll get to that.
To understand why Dojo was created, we have to first understand how AI works.
Basically, AI is a bunch of statistical models and algorithms implemented and trained on data so it can think and make decisions for itself. Data can be in any form, video, images, text corpus, or plain old numbers.
Now for AI to be efficient, it needs to be trained on a lot of data. In fact, it's a perfectly linear relationship, i.e., the more data AI is trained on, the more efficient and better it becomes.
Now for you to train AI on data, you need a lot of processing power; the more data you have, the more processing power you need, and for this purpose, AI engineers (like myself) make use of things called
But your data gets to a point where training them on GPUs are no more efficient, and at that point, your data becomes
To put this into perspective, Tesla, as a car/robotics😅company, makes use of AI, especially for their self-driving feature.
Tesla's AI self-driving feature is made of neural networks which are trained on millions of video data so it can become better (remember the more data, the better the AI), and these millions of video data easily become big data and so using GPUs would be inefficient and distributed computing would be great but to train AI on data as big as millions of data videos which are updated daily you'd need a lot of computers as essentially distributed computing is just sharing processing work between a lot of computers and so tesla decided that building a supercomputer would be more efficient both tech-wise and cost-wise and so the Dojo supercomputer was born😎.
Below is a picture of self-driving AI in action
To understand exactly how the Dojo supercomputer will work, you'd need to understand Tesla's data pipeline, which unfortunately is a trade secret😅(for some reason😒).
But from the little publicly available information, it goes something like this when you make use of AI, and it makes mistakes (it doesn't work as it should), it makes note of that mistake and improves on it.
Now let's take for instance, you own a Tesla (I don't😅), and you're driving it, and you're tired of taking the wheel and want to make use of the self-driving feature (I mean, it's just sitting there right), and it goes smoothly the self-driving AI is taking note of its surroundings using an inbuilt camera and sending that video back to Tesla's servers so that data can be used to train it more and in so that way the more people who use the self-driving AI, the more it gets better.
The cool thing also is the data collected when making use of the self-driving feature is also used to train the AI of new Tesla cars just being produced and so in essence, the AI just gets better and better, and that way, when you buy a new Tesla its self-driving will be better than the last one's.
Now back to the Dojo, in theory, when the video data gotten from the car is sent back to Tesla's servers, the Dojo supercomputer then gets hold of that data and uses it to train the neural network which powers the Tesla self-driving AI.
Another fun fact you should know is the data is being transmitted in real-time, so as a Tesla's self-driving feature is being used, the data is being transmitted, and so the Dojo supercomputer works full time.
The new Tesla Dojo supercomputer is very cool, and it helps improve the way AI is trained efficiently and it that way helps promote the way AI is used, and this is really good news for mankind.
AI is slowly integrating itself into our lives like Siri, Google assistant, Netflix movie recommendation, etc., and the Dojo supercomputer might well serve to improve the way our AI systems serve us.
Also Published Here