The Energy Cost of Execution-Idle in GPU Clusters
This addresses energy inefficiency in data centers for AI practitioners, offering incremental improvements to reduce GPU power waste.
The paper identifies execution-idle as a recurring high-power state in GPU clusters during low activity, accounting for 19.7% of in-execution time and 10.7% of energy across diverse workloads and GPU generations. It proposes prototypes for downscaling and load imbalance to mitigate this, with performance trade-offs.
GPUs are becoming a major contributor to data center power, yet unlike CPUs, they can remain at high power even when visible activity is near zero. We call this state execution-idle. Using per-second telemetry from a large academic AI cluster, we characterize execution-idle as a recurring low-activity yet high-power state in real deployments. Across diverse workloads and multiple GPU generations, it accounts for 19.7% of in-execution time and 10.7% of energy. This suggests a need to both reduce the cost of execution-idle and reduce exposure to it. We therefore build two prototypes: one uses automatic downscaling during execution-idle, and the other uses load imbalance to reduce exposure, both with performance trade-offs. These findings suggest that future energy-efficient GPU systems should treat execution-idle as a first-class operating state.