The rapid rise of generative AI is resulting in an unprecedented increase in data center energy consumption.
Data center power consumption could triple by 2030 to power the large AI models behind today's chatbots, image generators and more. This requires a coordinated response from your complete technology industry.
The recent NVIDIA GTC conference, some of the influential AI gatherings of the 12 months, featured a Taiwanese electronics company Delta Electronics discussed several cutting-edge advances to optimize energy consumption within the gigawatt data centers required to coach massive AI models.
Gigawatt data centers devour the electricity of tons of of hundreds of households yearly.
Delta's recent data center hardware includes modular racks, high-efficiency power racks and breakthrough voltage regulators mounted vertically on GPU boards to attenuate power loss. Solutions exhibited Exhibits on the GTC include:
- ORV3 server racks with high-density 48 kW power racks that increase efficiency by as much as 97.5%
- Vertical voltage regulator modules that reduce energy loss by 5-15% in comparison with traditional designs
- 800V DC power distribution with modular solid-state transformers for direct conversion of medium voltage grid power
- High-capacity monolithic UPS systems that deliver uninterrupted power with industry-leading 97.5% efficiency
Ralf Pieper, head of research and development at Delta's Custom Design Business Unit, said at NVIDIA GTC: “As the share of GPUs in AI computing increases, global data center power consumption is anticipated to greater than double and exceed 1 trillion kWh by 2026 amount.” ”
He continued: “Our unique expertise in high-efficiency server performance and DC/DC converters, in addition to ICT and energy infrastructure, enables us to advance the event of breakthrough solutions that may support the AI megatrend by optimizing the grid-to-chip. “Electricity Conversion Cycle.”
Delta estimates that its high-efficiency products have helped customers save nearly 40 billion kilowatt hours (kWh) since 200. For scale: 40 billion kWh could power about 3.73 million average American homes for a 12 months.
Combating the increasing resource demands of AI
Today's data centers already devour over 1% of the world's electricity. Microsoft, Google, Amazon, NVIDIA and others plan to expand their data infrastructure plans in the approaching years, with power consumption for hyper- and exascale data centers rising into the gigawatt range for the primary time in history.
Traditional nuclear fission power, nuclear mergerGeothermal and solar energy will all play their part in sustaining this – but increasing the efficiency of the AI infrastructure supported by Delta's technology is key.
Pieper concludes: “The introduction of gigawatt-scale data centers for AI training and inference applications, likely including 250kW racks, is inevitable. To ensure highly efficient, reliable and seamless power conversion and delivery to AI computing chips, Delta continues to develop progressive solutions for various layers inside the grid-to-chip ecosystem.”
