Home » AI-generated images require as much power as an entire smartphone battery

AI-generated images require as much power as an entire smartphone battery

by admin
AI-generated images require as much power as an entire smartphone battery

Today, when we ask ChatGPT/Dall-E, Stable Diffusion or Midjourney to generate the image we want, the processors in the cloud services are running hot in the background. Because AI-generated images require a lot of computing power and therefore a lot of energy. Researchers from the AI ​​startup Hugging Face and Carnegie Mellon University have now found out how much energy.

Conclusion of the study: Creating an image using a powerful AI model requires a similar amount of energy as fully charging a smartphone. The researchers also found that using an AI model to create text is significantly less energy intensive. According to the study, creating a text 1,000 times only uses as much energy as 16% of a full smartphone charge. This means that not only the computationally intensive training of AI models consumes a lot of energy, but also the daily use by hundreds of millions of people worldwide.

88 different models were tested using an NVIDIA GPU (A100-SXM4-80GB) hosted on Amazon Web Services and a proprietary tool called “Code Carbon” was used to measure power consumption. The name of the software already says what the researchers are trying to achieve. Because they also calculate: Generating 1,000 images with a powerful AI model like Stable Diffusion XL would cause approximately as much CO2 as driving 6.6 kilometers (4.1 miles) with an average gasoline engine.

Here is the crucial passage from the study:

„Charging the average smartphone requires 0.012 kWh of energy 4 , which means that the most efficient text generation model uses as much energy as 16% of a full smartphone charge for 1,000 inferences, whereas the least efficient image generation model uses as much energy as 950 smartphone charges (11.49 kWh), or nearly 1 charge per image generationalthough there is also a large variation between image generation models, depending on the size of image that they generate.“

Text generation is twice as energy hungry as Google search queries

Generating text produces much less CO2 – the least carbon-intensive text generation model examined was only responsible for as much CO2 as driving 0.0006 miles – the equivalent of just under a meter. For comparison, Google once estimated that an average online search query uses 0.3 watt-hours of electricity, which is equivalent to driving 0.0003 miles, or half a meter. This means that generating texts using AI models is twice as CO2-intensive as a Google search.

See also  Bloody Los Angeles!Open world action-adventure game "Dead Island 2" is available today "Dead Island 2" - Bahamut

“We found that general-purpose generative architectures are orders of magnitude more expensive than task-specific systems, even when taking into account the number of model parameters,” the researchers write. They point out that the benefits of these AI models should be consciously weighed against the increased costs in the form of energy and emissions – for example if you let Dall-E generate images (“Make it More”) just for fun to follow a fast-moving trend.

It is well known that AI models have major deficits in energy consumption. For providers such as OpenAI, Google and Google, the question now arises as to how they can make the models more efficient so that they have to pump in less computing power – and also how and whether they can operate the GPUs with renewable energy.

AI Foundation Models have major deficits in copyright & energy consumption

You may also like

Leave a Comment

This site uses Akismet to reduce spam. Learn how your comment data is processed.

This website uses cookies to improve your experience. We'll assume you're ok with this, but you can opt-out if you wish. Accept Read More

Privacy & Cookies Policy