Tesla uses 10,000 H100 GPUs in its $300 million supercomputer. This processor is for artificial intelligence.
Tesla is trying to speed up the development of the FSD driver assistance system by boosting its supercomputer. Apparently, Tesla’s $300 million supercomputer uses 10,000 Nvidia H100 GPUs. Nvidia has made this chip specially for artificial intelligence processing.
According to an analyst in the X social network (formerly Twitter), Tesla has been using Nvidia’s A100 chip, and thanks to the H100, its supercomputer achieves about five times more processing power (up to 340 petaflops in FP64 format).
Elon Musk attaches great importance to FSD and says that the future of Tesla is tied to this technology. Tesla has been working on optimizing FSD for years, but the driver-assist system has yet to leave beta mode. Tesla is building a $1 billion supercomputer called Dojo by 2024 to speed up FSD development.
Tesla wants to invest more than four billion dollars in the development of FSD and the development of artificial intelligence models. The purchase of 10,000 H100 graphics chips brings the company closer to its goals.
The H100 is currently the most requested product in the technology industry, and the supply of 10,000 units is a significant achievement for Tesla. Elon Musk previously said that buying artificial intelligence chips is harder than buying drugs.
The price of Nvidia H100 graphics is between $25,000 and $30,000, and the green team apparently spends $3,200 to produce these graphics. Nvidia is the only company in the world that can make a processor of the size of H100.