How to fix the AI energy crisis

 https://www.nature.com/articles/d41586-024-03408-z?mc_cid=eaa1a99559&mc_eid=7f01981b78


The data centres needed to power AI are guzzling electricity by the gigajoule. By 2026, the International Energy Agency predicts data centres’ energy consumption will increase by between 35% and 128%, adding each year something between the annual energy consumption of Sweden and Germany. Potential remedies for this looming energy crisis include introducing new chip architectures, switching to analogue computing and using photonics to encode data in light instead of wires.


At this point of potential crisis, many hardware designers see an opportunity to remake the basic blueprint of computer chips to make them more energy efficient. Doing so will not only help AI to work more efficiently in data centres, but also enable more AI tasks to be carried out directly on personal devices, for which battery life is often crucial. However, to convince industry to embrace such large architectural changes, researchers will have to show significant benefits.

According to the International Energy Agency (IEA), in 2022, data centres consumed 1.65 billion gigajoules of electricity — about 2% of global demand. Widespread deployment of AI will only increase electricity use. By 2026, the agency projects that data centres’ energy consumption will have increased by between 35% and 128% — amounts equivalent to adding the annual energy consumption of Sweden at the lower estimate or Germany at the top end.

 Google’s 2024 environmental report revealed its carbon emissions have increased by 48% in 5 years. In May, Microsoft president Brad Smith in Redmond, Washington, said that the company’s emissions had risen by 30% since 2020. 

The high energy consumption associated with training and operating AI models is due in large part to the reliance of these models on huge databases, and the cost of moving those data between computing and memory, in and between chips. When training a large AI model, up to 90% of the energy is spent accessing memory.

In 2023, IBM described1 an early analogue AI chip that could perform matrix multiplication at 12.4 TOPS per watt.


Popular posts from this blog

Five Sustainability Heroes New Electronics

 https://www.newelectronics.co.uk/content/features/five-sustainability-heroes  Conclusion Increasing sustainability is not just a single significant action, it can also be the combination of thousands of smaller changes that come together to form a greater net change. Through smarter component selection, deeper development and intelligent maintenance techniques, the electronics industry can work to reduce energy consumption, as well as lower raw material requirement through greater integration, maintenance schedules, and miniaturised designs. The sustainability ‘heroes’ highlighted here are a diverse range of products and solutions that can provide improvement through both significant innovation and through continued iterative improvement. Such components and systems can be used by engineers today to help improve the sustainability of future designs.
Read more

How ‘green chips’ are powering a sustainable future

Image
 https://datacentrereview.com/2021/12/how-green-chips-are-powering-a-sustainable-future/ The link between the exponentially growing amount of data and the rising global temperature is direct and proven. Processing the ever-increasing amount of data requires immense energy — and the energy sector is responsible for  approximately 72%  of the world’s greenhouse emissions, making electricity the largest contributor to global warming, according to Greenpeace. Just like a smartphone gets hot when the battery or CPU overloads, a data centre gets hot as the amount of data traffic increases. Cooling it off requires substantial amounts of energy, which is why quite a few data centres are in cold regions like Finland, Sweden and Iceland, where cold air and seawater are in abundance. The future is looking challenging, with the amount of energy used to store and process human data forecast to surpass the amount of energy used to produce it.  IDC’s Global DataSphere  estimat...
Read more

Overcoming Changes and Challenges with Sourceability’s Datalynq

https://datalynq.com/post/how-to-manage-the-complexity-of-market-availability-for-electronic-components (Abril 2024). Datalynq is Sourceability’s market intelligence tool. It was recently integrated into Cadence’s OrCAD X Platform under Live BOM to provide design engineers with early warnings about risk-prone components in their designs. By replacing these components early with safer parts, product designs become more sustainable in the long run, preventing problems with advanced and mature nodes.  
Read more