Mark Zuckerberg shared some interesting information about the development of Meta’s artificial intelligence infrastructure, which will include a surprising number of NVIDIA H100 Tensor GPUs. In his post on Instagram, Meta’s CEO noted the following: “We are currently training the next-generation Llama 3 model, and building massive computing infrastructure to support our future plans, including 350k H100s by the end of this year – And a total of almost 600k H100s equivalent to computing if you include other GPUs”.
This means that the company will boost the AI infrastructure with 350,000 Nvidia H100 GPUs on top of the existing GPUs, which reach the number of 250,000 (again H100). So the total The number of Nvidia’s H100 GPUs that Meta will have at the end of the year is estimated at 600,000, a huge number by general consensus.
The number, however, is expected to come at a high price. With the average selling price of a H100 GPU approaching 30,000 dollars, Meta’s investment will amount to an astronomical amount of 10.5 billion dollars. There should be other GPUs in the Meta infrastructure, but most will include the NVIDIA Hopper family. In addition, Meta is currently training the LLama 3 AI model, which will be much more capable than the existing LLama 2 family and will include better reasoning, coding and mathematical problem solving capabilities. These models, as the company has specified, will be open source and will appear on GitHub.
Later, as artificial general intelligence (AGI) is expected to come into play, Zuckerberg notes that “our long-term vision is to build general intelligence, open it responsibly, and make it widely available so everyone can benefit.” ».
0 Comments