Artificial intelligence (AI) is transforming industries, driving innovation and efficiency. However, sustainability is crucial for long-term benefits.
Sustainable AI growth faces challenges, including energy consumption. Large language models (LLMs) require significant computational power, leading to high energy consumption.
Recent comments from industry executives anticipate a factor of 100 times more computes required for the next generation of LLM training, while the energy needed to train a leading model is multiplied by 10 every two years.
Processing of LLMs is typically partitioned over different accelerators to optimize performance, driving key parameters needs.
Author's summary: ST silicon photonics and BiCMOS technologies enable sustainable AI growth.