When you hear the word “Tesla”, what pops up in your mind? Chances are, you are thinking about electric vehicles that are gradually replacing the traditional motorized ones. While you are not wrong, Tesla has more tricks up its sleeve that you might not even know existed. Take their recently unveiled Super Computer, for example.
Being used to train their self-driving AI, this thing is already the world’s 5th most powerful supercomputer that was ever created! On top of that, it seems like they are just getting started as more upgrades are on their way. But why is this supercomputer so much important for tesla? What’s it capable of? Let’s find out.
Overview of Supercomputers
Before talking about everything else, let us go through how supercomputers evolved over the decade. Released in 1694, the CDC 6600 is considered to be the world’s first supercomputer that was developed and manufactured by Control Data Corporation. It was used for high-end mathematical and scientific computing with a power of 3 million floating points executions per second.
Fast forward to recent times, the Dojo supercomputer from Tesla is claimed to have a speed of about 362 teraFLOPS, which can do 362 trillion floating-point executions per second. So we get the idea of how much performance escalation has occurred since the release of the CDC 6600. But what is the reason behind Tesla’s sudden interest in supercomputing?
Why Do Tesla Supercomputers Exist?
Well, as it turns out, Tesla is using their in-house supercomputer to train and provide power to neural nets and self-driving AI that are used in Tesla’s Autopilot vehicles. Well, this makes sense as they are focusing on computing power to be used on the internal and external aspects of their self-driving vehicles. The custom-built supercomputer is expected to elevate Tesla’s capacity to train neural nets using video data, which is critical to its computer vision technology powering its self-driving effort.
“Supercomputers of this caliber are much needed for Tesla as Elon Musk has the plan to achieve Level 5 of Autonomy in their self-driving vehicles.” Now, if you don’t know how the level of Autonomy works in self-driving vehicles, let us give you an idea. There are five levels of driving autonomy. On Level 1, the driver gets to enjoy at least one support system for an assisted driving experience. On level 2, the system can take control of the steering, braking, and acceleration whenever it is needed. Level 3 is a significant leap over Level 2 automation as it uses multiple driving assistance and AI to drive and control the car on its own. On level 4, low to no human interaction is required. And lastly, on level 5, you can basically sleep in the car as the car takes you to places on its own. While it may sound like something straight out of a sci-fi movie, Tesla hopes to achieve that with the help of its supercomputers.
Important Specs of Tesla Supercomputer
With that being said, let us take a look inside this Tesla supercomputer and see what is powering this beast. Under the hood, it packs 720 nodes of 8x A100 80GB, which is equivalent to 5760 GPUs in total. With a processing power of 1.8 Exa Flops, this thing also has 10 Petabytes of NVMe storage that can run at 1.6 terabytes per second. And lastly, the 640 Terrabytes per second switching capacity on this thing is rigged to execute massive data processing and AI training at its best.
In addition to that, Tesla is also working on their D1 Dojo chip that will be integrated into its upcoming supercomputers. Introduced in their recent AI Day event, this super-performing chip has a power of 362 TFLOPS, 354 Training Nodes, and 1.25MB of SRAM per unit coupled with a huge bandwidth of up to 512 GB per second. Which is unlike something we have ever seen before.
The Future
While Elon made quite an advancement with Tesla’s supercomputer, it seems like he is just getting started. In a recent Tweet, he mentioned being “Working on Master Plan Part 3” which hints toward broad-scale Tesla operations to shift humanity away from fossil fuels.
Other than that, notateslaapp predicts that Master Plan Part 3 might also include an even better AI training supercomputer that stretches further beyond self-driving cars. Andrej Karpathy, Tesla’s head of AI, also revealed their roadmap regarding the Dojo supercomputer, where they are expecting about 10 times the performance improvement on their next generation of supercomputers. And if that turns out to be true, we are about to see a greater shift in humanity’s advancement that has the potential to change for the greater good.
