Generative AI is accelerating data centres' demand for electricity globally. By 2026, total energy consumption is projected to reach more than 1,000 TWh alone. According to Morgan Stanley, generative AI could use as much energy as Spain needed to power itself in 2022.
So, how do we cater to this growing demand in an environmentally friendly way? As an organisation exploring AI initiatives, what sustainability challenges should I be aware of, and what should I look for in my AI compute provider?
As the demand for AI solutions grows, so does the environmental footprint. Training large-scale AI models require significant computing resources, leading to increased energy consumption and higher carbon emissions. For instance, a study by the University of Massachusetts Amherst found that training a single AI model can emit as much carbon dioxide as five cars over their lifetimes, totalling approximately 626,000 pounds of CO2.
The surge in AI applications has created an unprecedented demand for infrastructure, putting immense pressure on power grids worldwide. Data centres already account for nearly 2-4% of global CO2 emissions, comparable to the aviation industry. This often leads to restrictions on new data centre developments due to their environmental impact.
Training large-scale AI systems is both power and water-intensive. For example, Sasha Luccioni at Hugging Face notes that generative AI systems can consume 33 times more energy than machines running task-specific software. Additionally, Bluefield Research reported that training ChatGPT-3 required over a million gallons of water. Training GPT-3 consumed an estimated 1.287 gigawatt-hours (GWh) of electricity, resulting in significant carbon emissions.
Choose a provider that prioritises energy-efficient technologies and practices. Look for data centres that utilise advanced cooling methods, such as natural cooling in colder climates, to reduce energy consumption.
Ensure your provider relies on renewable energy sources. Data centres powered by 100% renewable energy, like those near the Arctic Circle in Norway, significantly reduce the carbon footprint.
Select data centres in regions with an abundant renewable energy source. This ensures a stable energy supply and prevents additional strain on existing power grids.
Opt for providers that offer vertically integrated platforms encompassing data centre ownership, hardware, and software. This integration ensures every layer of the AI stack is optimised for efficiency and performance, contributing to a favourable Green ROI.
A good AI compute provider should offer flexible AI cloud platforms that simplify the journey from development to production. This helps organisations scale their AI initiatives efficiently while working towards decarbonisation goals.
Nscale’s sustainable data centre provides a solution for achieving environmental and financial goals. Our data centres are strategically located near the Arctic Circle in Norway, benefiting from the cold climate for more efficient cooling, powered by 100% renewable energy.
By targeting locations with an oversupply of renewable energy, Nscale ensures that the progression of AI and HPC data centres does not strain existing power grids, providing sustainable and efficient operations. Deploying AI infrastructure clusters in these areas allows us to utilise energy otherwise wasted.
Nscale’s vertically integrated platform ensures every layer of the AI stack is optimised for efficiency and performance. Our flexible AI cloud platform simplifies the journey from development to production, helping organisations achieve their AI and decarbonisation goals.
Seamlessly transition your LLM training from Nvidia to AMD GPUs without altering your core training logic.
.png)
Optimising AI model performance: vLLM throughput and latency benchmarks and GEMM Tuning with rocBLAS and hipBLASlt
Unlock the full potential of GPUs with Nscale and AMD in our upcoming webinar on the 30th of May 2024 at 3 pm BST.
