RAVEL and Sandia test power-aware AI workload orchestration

Image
RAVEL and Sandia test power-aware AI workload orchestration

SEATTLE -- November 12, 2026 -- RAVEL and Sandia National Laboratories have launched Project Quicksilver, a multi-phase study of power-aware AI workload orchestration across geographically distributed computing sites. Phase One will run on three Sandia locations linked by a wide-area network, and it compares air-cooled and immersion-cooled environments.

Phase One tests two cooling types at three Sandia sites

The work runs under a Cooperative Research and Development Agreement. It uses RAVEL Orchestrate AI, which the company describes as a federated layer that schedules workloads across heterogeneous, distributed resources using power, cost and performance inputs. Sandia is the initial research lead.

Researchers will use Box-Behnken and Plackett-Burman experimental designs and will vary network bandwidth and speed alongside other factors. They will run identical workloads in each cooling environment and measure three things:

  • Energy efficiency: energy use and operating costs, including TUE and PUE-adjusted efficiency and potential heat-reuse benefits.
  • Total cost of ownership: how scheduling and placement decisions change CAPEX and OPEX for infrastructure, power, cooling, networking and operations.
  • Workload time-to-completion: how placement decisions affect AI workload execution time.

The design pairs cost and speed in one test. Operators weighing where to place AI capacity would get measured trade-offs between energy, spending and execution time, rather than vendor estimates. The release describes no commercial deployment. It says the project aims to produce evidence that could inform future data center design and operations.

Sandia plans to bring in more Department of Energy laboratories

Sandia plans to expand Project Quicksilver to additional U.S. Department of Energy national laboratories and to explore further use cases for environmental and performance optimization. The companies did not disclose the project's duration, funding or the start date of Phase One.

RAVEL CEO and co-founder Denise Muyco says the next phase of AI infrastructure leadership will depend on how efficiently operators use every megawatt, GPU and dollar. Dave Martinez, engineering program and project lead for Infrastructure Computing Services at Sandia, says the lab will explore whether dynamic workload placement can make better use of available power and computing resources while balancing performance and cost.

RAVEL and Sandia present the test plan at SC26 on November 17

The two will present Project Quicksilver at SC26, which runs November 15-20, 2026. The session, "Project Quicksilver: Dynamic Power-Aware AI Orchestration Across Distributed Computing Environments," is scheduled for 2:30 p.m. CST on Tuesday, November 17, at the U.S. Department of Energy booth. That is the third day of the six-day conference, by FairsOnline's calculation.

According to the companies, the presentation will cover the experimental approach, planned test phases and anticipated research outputs.

Published by:
fairsonline_team
Industries
News Type