NVIDIA, ETAP and Schneider Electric: Developing Digital Twins for AI Factories
In an effort to meet the challenges of power management and energy efficiency in the era of power-hungry artificial intelligence training, three industry giants are teaming up to solve those issues by digital twin technology.
Energy management firm Schneider Electric announced Tuesday it is working with power system designer ETAP and chipmaker NVIDIA to develop digital twins which can simulate the power needs of AI-enabled factories at a deeper, more minute level. The technology will leverage NVIDIA’s Omniverse Blueprint framework.
Whereas basic visualization of electrical systems has long been possible, the integration of ETAP and NVIDIA Omiverse technologies are teaming to create a comprehensive AI Factory digital twin. ETAP will provide modeling technology to create a virtual replica of a data center’s electrical infrastructure and combine it with real-time power system data and analytics.
The rise of AI training models is driving already high data center energy demand to unprecedented levels of consumption, threatening both grid resource adequacy and on-site power performance. Digital twin technology can increase precise modeling of dynamic load behavior, the companies say.
“As AI workloads grow in complexity and scale, precise power management is critical to ensuring efficiency, reliability and sustainability,” Dion Harris, senior director of HPC and AI Factory Solutions at NVIDIA, said in a statement. “Through our collaboration with ETAP and Schneider Electric, we’re offering data center operators unprecedented visibility and control over power dynamics.”
Earlier this year, a report by Goldman Sachs forecast that AI training will increase data center power demand by 165% by 2030. Hyperscale cloud-based data centers and other companies are building large language models for AI training which include vast amounts of information using power-intensive processors, reads the Goldman Sachs report.
ETAP and NVIDIA are deploying a new "Grid to Chip" approach which is aimed at solving power management and optimization challenges in AI computation. Currently, data center operators estimate average power consumption at the rack level, but the new digital twin aims to model dynamic load behavior at the chip level to improve power system design and energy efficiency.
"Collaboration, speed and innovation are the driving forces behind the digital infrastructure transformation that's required to accomodate AI workloads," said Pankaj Sharma, executive vice president for data centers, network and services at Schneider Electric.
