Nvidia uses AI to design faster chips

Designing a GPU is labor-intensive and time-consuming. Bryan Catanzaro, Nvidia’s vice president of applied deep learning research, says a single chip takes nearly 1,000 people to build, and each person needs to understand how the different parts of the design work together.

The ChipNeMo system uses a large language model developed from Meta's Llama 2. According to Insider , ChipNeMo's chatbot can answer queries related to chip design, such as GPU architecture and write chip design code.

Nvidia is a beneficiary of the AI ​​craze

In 2023, the AI ​​craze pushed Nvidia into the “trillion-dollar club,” with a market capitalization of $1 trillion. Goldman Sachs analysts expect Nvidia’s stock to continue rising through the first half of 2025.

Since ChipNeMo launched in October 2023, Nvidia says the AI ​​system has been useful in summarizing notes and training new chip design engineers. The company is working to ramp up production to meet surging demand for chips.

In January, Mark Zuckerberg announced plans to spend billions of dollars to buy 350,000 more Nvidia H100 GPUs to fuel the AI ​​race. If you add in other models, Meta will have amassed 600,000 chips by the end of 2024.

Several other tech giants are also looking to address the chip shortage.

In July 2023, Google’s DeepMind division created an AI system to speed up the design of its latest custom chip, according to the Wall Street Journal . Meanwhile, leading chip design company Synopsys launched an AI tool designed to help chip engineers increase productivity.

Source link: https://thanhnien.vn/nvidia-dung-ai-de-thiet-ke-chip-nhanh-hon-185240209032238931.htm