Future of Autonomous Driving

Gigantic Fans for Tesla's Data Center

Tesla is setting new standards: gigantic fans are currently being installed for Giga Texas to cool the new data center and the 50,000 Nvidia GPUs needed to train the Full Self-Driving (FSD) system.

By Thomas Schröter

19.06.2024, 21:08•3 min. Reading time

Tesla's new data center in Austin, Texas, marks a significant step towards advanced AI development and autonomous driving. In a YouTube video released on June 18, 2024, you can see how Tesla's new data center in Austin, Texas, is being equipped with cooling towers to dissipate the heat from 50,000 Nvidia H100 GPUs with a total capacity of 35 MW.

Cooling and Performance at the Highest Level

The construction of the data center is progressing rapidly. The installation of the gigantic fans is necessary to handle the enormous heat dissipation of 35 MW. This capacity is required to effectively dissipate the heat from 50,000 H100 Nvidia GPUs. Tesla plans to operate up to 85,000 of these powerful GPUs for FSD training purposes in the future.

Investment in Hardware

The list price of a single H100 Nvidia GPU is 40,000 euros. Calculated for the planned 85,000 GPUs, this results in a theoretical total value of around 3.9 billion euros. However, it is likely that Tesla will pay significantly less due to the bulk purchase price. This substantial investment demonstrates Tesla's commitment to advancing autonomous driving development.

Improved Computing Power

A key aspect of the new data center is the computation of the FSD software for the latest hardware version 4 (HW4). Currently, the FSD software is still running in HW3 emulation mode. With the new data center, an increase in computing power from 36 TOPS (Tera Operations per Second) to 50 TOPS is targeted, which will massively improve the efficiency and accuracy of autonomous driving.

Technical Data Comparison

HW3 (& FSD Computer 1)	HW4 (& FSD Computer 2)
Samsung Exynos-IP based	Samsung Exynos-IP based
12 CPU cores	20 CPU cores
CPU frequency: 2.2 GHz	CPU frequency: 2.35 GHz
2 Neural Network (NN) processors	3 Neural Network (NN) processors
NN processor frequency: 2.0 GHz	NN processor frequency: 2.2 GHz
14 nm process	Estimated TSMC’s 7nm or N4 (4nm class)
36 TOPS	50 TOPS
1.2MP cameras around the car	5MP cameras around the car
Previously used radar was Continental’s ARS410 with a range of 160 meters. Discontinued in 2021.	Speculated to be the Arbe Phoenix radar unit, with greater detail and a range of 300 meters.

Comprehensive Training Material

Tesla has access to a vast archive of real driving videos from 6 million vehicles, which are used as training material for the FSD system. Additionally, the so-called "Shadow Mode" allows for the unnoticed testing of new versions of the AI approach, collecting an immense amount of supervised miles. These data are crucial for the continuous improvement of autonomous driving.

Data is the cornerstone in the race to fully autonomous vehicles. Real driving examples are used to train neural networks. Every Tesla vehicle is equipped with an FSD computer, running two FSD systems simultaneously. The first system is for driving and control in autopilot mode, while the second FSD system constantly runs in shadow mode. The shadow mode is a simulation. If the driver reacts differently than the simulation or the FSD does not know what to do, it is recorded as an inaccuracy. Tesla can then collect data with similar inaccuracies and retrain the neural networks to better respond to scenarios in the future.

Regulatory Challenges and Market Position

The European regulatory authority recently tested Tesla's FSD software, which could indicate future legal challenges and adjustments. These tests are of great importance as they form the basis for the approval of FSD technology in Europe.

German Tesla driver experiences FSD in the USA for the first time

With the planned computing power, Tesla will rank among the top 10 most powerful data centers in the world. This not only underscores Tesla's technological leadership but also the enormous potential of the company in the field of AI and autonomous driving.

Tesla's solar power system on the roof of the Gigafactory in Austin will deliver 30 MW of power. This means that the data center can be powered by this renewable energy source for a quarter of its needs. This demonstrates Tesla's commitment to sustainable energy supply and environmentally friendly technologies.

Audi and BMW are also investing heavily in autonomous driving. The European regulatory authorities are testing and preparing the legal framework for autonomous driving.

Tesla has established itself as a leader in autonomous driving and AI. The new data center in Austin will further strengthen this position and provide the foundation for future developments in this field.

With these developments, Tesla demonstrates that the company is not only a pioneer in electric mobility but also plays a leading role in autonomous driving and AI development. The future looks bright for Tesla and the progress of autonomous driving.