In partnership with Supermicro, US Virtual Power Producer (VPP) CPower Energy and AI orchestration platform Bentaus have completed a joint demonstration project on improving AI compute's relationship with the grid.
As part of the test, CPower dispatched real-time wholesale electricity market signals from California Independent System Operator (CAISO) through the Bentaus energy orchestration platform to Supermicro-managed GPU infrastructure.
According to the companies, the GPU infrastructure responded to a full dispatch signal cycle in less than 20 milliseconds.
In addition, the companies said that the cluster of Supermicro servers with Nvidia B200 GPUs included in the test maintained AI workloads and service-level agreements while using up to 75 percent less electricity when responding to grid conditions.
"The ability to use AI compute loads to help the grid keep up with rapid load growth has been largely unrealized until now. This successful end-to-end test is a breakthrough, accelerating flexible compute's emergence as a controllable, grid-interactive load, and moving it from the theoretical to the actual," said Michael D. Smith, CEO, CPower.
The partners claim that the test established a foundational architecture for integrating AI compute into demand response through validating abilities such as translating energy market signals into compute-level orchestration actions, dynamically curtailing and restoring AI workloads across GPU clusters, and coordinating compute control with infrastructure lifecycle management.
The partners have said that following the test, they will collaborate with data center operators and market participants across multiple US wholesale power markets.
"Demonstrating that GPU workloads can react to market signals in the blink of an eye is an important step toward aligning the growth of AI with the operational realities of the grid," said Robert Davidoff, CEO, Bentaus.
"AI factories can evolve from passive consumers of electricity into active participants in wholesale power markets."
Bentaus' orchestration platform works by integrating IoT sensors and Edge AI across all platforms to monitor and manage energy, carbon, and operational metrics in real-time. The company also offers an optimized energy management service to optimize energy consumption for large power consumers such as data centers.
The use of orchestration software within AI data centers has seen increased uptake recently. Nvidia-backed EmeraldAI has been at the forefront of this through its Emerald Conductor orchestration platform. The company has run a number of demonstration projects in Phoenix, Chicago, and, most recently, in the UK, in partnership with National Grid.
Comments