Meta Engineer Claims Only Two Additional Nuclear Power Plants Needed to Meet Global AI Inference Energy Demand

Recently, Sergey Edunov, Director of Generative AI Engineering at Meta, revealed at the Silicon Valley Digital Workers Forum that to meet the growing demand for AI application inference globally next year, the power generated by just two additional nuclear power plants would suffice. Edunov estimated that a significant number of Nvidia H100 graphics processors will be added globally for AI inference next year, and if all were used for the generation of reasonably scaled language models, the power consumption would be within manageable limits. He stated that training larger-scale language models would face data scarcity issues, and whether current technology can achieve general AI may be known within the next 3-4 years.

Taobao's 100 Billion Subsidy Launches a Special Session for Farming Prawns Hardware: Mac mini in Stock Subsidized to 3999 Yuan

With the popularity of AI applications for farming prawns, the demand for terminal computing power has surged. Apple's Mac mini has become popular due to its high energy efficiency, leading to a temporary shortage and price premium in the market. Taobao's 100 Billion Subsidy has launched a special session to stabilize prices through official subsidies. Among them, the Mac mini equipped with M4 chip is now available at 3999 Yuan, offering significant discounts.

Researchers Discover an Energy-Efficient Method for Training Large Language Models, Reducing Energy Consumption by 30%

Recently, a new study from the University of Michigan found that an energy-efficient method for training large language models can achieve similar results in the same amount of time while reducing energy consumption by 30%. This method can save enough energy to power 1.1 million American households by 2026. The researchers developed a software tool named Perseus that identifies the critical path, which is a series of subtasks requiring the longest time to complete. Then, Perseus reduces the processor speed on non-critical paths so that all tasks can finish simultaneously.

The World's Fastest AI Inference Service Is Here! 20x Speed Increase at Extremely Low Cost

Cerebras Systems has launched Cerebras Inference, claiming it to be the fastest AI inference service in the world, achieving performance that surpasses traditional GPU-based systems by 20 times with significantly improved cost-effectiveness, particularly suitable for processing large language models (LLMs). Its 8B version processes 1800 tokens per second, while the 70B version processes 450 tokens, with speed and cost-performance far exceeding NVIDIA GPU solutions.

The Energy Consumed to Generate an AI Image is Equivalent to Charging a Mobile Phone

Using AI to generate images, write emails, or ask chatbots contributes a certain burden to the planet. The energy consumed to generate an image using AI is equivalent to charging a mobile phone fully. The energy consumption for generating text is lower, as generating text 1000 times only consumes 16% of a mobile phone's charging capacity. The use of large generative AI models is more energy-intensive than smaller models.

Meta Engineer Claims Only Two Additional Nuclear Power Plants Needed to Meet Global AI Inference Energy Demand

Related Recommendations

Taobao's 100 Billion Subsidy Launches a Special Session for Farming Prawns Hardware: Mac mini in Stock Subsidized to 3999 Yuan

Unveiling the Energy Consumption of Google Gemini! A Single Response Only Uses One Second of a Microwave's Power, the AI Green Revolution is Quietly Beginning

Researchers Discover an Energy-Efficient Method for Training Large Language Models, Reducing Energy Consumption by 30%

The World's Fastest AI Inference Service Is Here! 20x Speed Increase at Extremely Low Cost

The Energy Consumed to Generate an AI Image is Equivalent to Charging a Mobile Phone