Top Tweets for #Trsla

As a big fan of Elon Musk and Tesla, I've been closely following Tesla's AI progress. A defining moment last year was when Elon said he was still buying as many H100s as possible, indicating that Tesla is constrained by computing power. I tried to decode what this means and reasoned it out using first principles. I learned that despite spending billions of dollars and several years, Tesla's Dojo is still far from being able to replace Nvidia's A100 & H100s, suggesting that Nvidia must have done a lot of things right. This made me think that having a fully vertically integrated system is good for Tesla when it can move faster than the industry, but it can be detrimental if it's slower (i.e., Dojo < H100s) or if the chosen solution is incorrect (e.g., whether relying solely on vision is right). One advantage Tesla has over others is its access to real-world data. But, thinking from first principles, this can be solved through simulation, or considering that the industry collectively has more cars than Tesla alone. I concluded that Tesla's Full Self-Driving (FSD) doesn't have a 100% success rate, but Nvidia does. No matter the direction of the industry, they have something valuable to offer. Let's look at Nvidia's autonomous platform. Nvidia's DRIVE® end-to-end computing full-stack platform for autonomous vehicles. Enable anyone to build anything autonomous In-car solution DRIVE Hyperion is the in-car solution, a vehicle architecture that includes sensors, DRIVE AGX for compute, and software tools necessary for robust self-driving and intelligent cockpit capabilities. In the data center NVIDIA provides the hardware and software you need for AV development, including NVIDIA DGX™ to train DNN for perception and DRIVE Sim for generating data sets and validating the entire AV stack. Open and flexible It offers various levels of engagement for customers. They can choose to purchase just the chips, or chips along with algorithms. There's also the option to use the operating system or the Omniverse virtual world simulator for synthetic data generation and testing robots in a virtual environment. This openness has led to customers working with the company across these different areas. $NVDA $TSLA