Carlo Ruiz, Senior Director, Data Centre Solutions EMEA at NVIDIA, explains why it’s vital to prioritise sustainable computing initiatives now to meet net zero targets in the future.
This summer, yet more climate records were broken in the UK. For the first time since records began, temperatures in September exceeded 30°C for four consecutive days. In the fight against climate change, technology is one of our most important allies and we must prioritise ensuring that innovation does not come at the cost of the environment.
Sustainable computing initiatives are integral to reaching net zero climate targets and extracting maximum energy efficiency across the entire computing supply chain. While many companies are already integrating green computing solutions into their practices, there is still significant work to be done across the board to prioritise sustainable computing solutions.
Getting ahead of the curve
Sustainable computing is key to ensuring we can use technology to fight the climate crisis effectively. Also known as green computing, the term refers to minimising the negative effect computing technologies have on the environment. The term is holistic in its understanding, meaning that sustainable computing initiatives do not stop at simply the energy efficiency of the device in your pocket, the computer at your desk or the supercomputers powering the world’s most powerful AI tools. Rather, it translates to the entire supply chain. This means from design and production to the eventual recycling of the product, a sustainability-first mindset needs to be present at each step of the journey.
Global electricity consumption increases by approximately 3.4% every year. Data centres only account for a small percentage – around 2% – of the 26,000 terawatt-hours consumed in 2022, but that percentage could rise as high as 8% by 2030. By making sustainable computing solutions second nature today, we will be better equipped to combat the consequent impacts of growing demand tomorrow.
How can AI play its part?
AI tools are being used by climate scientists and companies alike to accelerate sustainability efforts. It is no surprise that these intensive tasks require significant resources, but the costs certainly don’t outweigh the benefits.
The supercomputers behind large simulations and AI tools can be fine-tuned for efficiency by utilising appropriate hardware and software. GPUs (graphics processing units) and DPUs (data processing units) facilitate increased efficiency for such high-performance computing tasks. GPUs process workloads in parallel, meaning more can be achieved in less time and with a decreased demand on energy.
Performance data shows that if every CPU-only server running AI and high-performance computing worldwide switched to a GPU-accelerated system, around 20 trillion watt-hours of energy could be saved per year. That’s equivalent to the electricity requirements of nearly 2 million U.S. homes.
The United States’ National Energy Research Scientific Computing Centre (NERSC) recently tested efficiencies across four of its key high-performance computing and AI applications. The NERSC recorded how quickly applications ran, as well as how much energy was consumed by one of the world’s largest supercomputers when running on CPU-only and GPU-accelerated nodes. The test found that when using GPUs, energy efficiency rose, on average, by five times. Evidently, as well as reducing the environmental impact, reducing the amount of energy required also drives down costs.
The benefits to improving efficiency are not only found in optimising hardware. Similarly, optimising an AI model’s software can result in significant benefits. For example, Colossal-AI’s Gemini enables more efficient management of memory space by using GPU and CPU memory at the same time. This means large AI models can be trained using just one GPU, allowing AI developers at universities and small companies to train models much more efficiently.
Simulating with digital twins
Companies like Lockheed Martin, the global security and aerospace company, have already started using AI tools directly to combat climate change. Through the use of digital twins, digitally created physically accurate replications of environments and objects, the company has been able to simulate the spread of forest fires and better understand how to fight them effectively.
The value of these tools cannot be understated, especially as the UN is forecasting that the occurrence of extreme fires could increase by up to 14% by 2030 and 30% by 2050. Unlike standard simulations, digital twins provide enhanced realism and allow for much more accurate predictions. In examples such as this, where trying to predict the unpredictable can both save lives and our planet, the value of increased accuracy is clear.
This isn’t the only way that AI tools are being used to combat forest fires. As well as utilising AI in simulations, AI-powered cameras and sensors are being deployed throughout California to provide timely alerts to first responders fighting fires. These technologies detect and alert authorities to the early signs of fires which enables quicker response times and increases the chances of managing devastating fires. Both in simulations and on the ground, AI tools are being deployed to reduce the impacts of forest fires.
Computing technology is a vital tool in the fight against climate change. But these technologies must also be sustainable if they are to have the greatest impact. To reach net zero targets, every industry must play its part. For computing, this means prioritising the energy efficiency of computers, assessing the environmental impact of the entire supply chain and using these powerful technologies to advance the work of climate scientists around the world.