Plenty of Room in the Middle: Revealing Opportunities on a GPU Cluster via Fine-Grained Monitoring

Efficient utilization of GPU resources and power has become critical with the growing demand for GPUs in high-performance computing (HPC). In this paper, we analyze GPU utilization and GPU memory utilization, as well as the power consumption of the Vienna ab initio Simulation Package (VASP), using the Slurm workload manager historical logs and GPU performance metrics collected by NVIDIA's Data Center GPU Manager (DCGM). VASP is a widely used materials science application on Perlmutter at NERSC, an HPE Cray EX system based on NVIDIA A100 GPUs. Using our insights from the resource utilization analysis of VASP applications, we propose a resource prediction framework to predict the average GPU power, maximum GPU utilization, and maximum GPU memory utilization values of heterogeneous HPC system applications to enable more efficient scheduling decisions and power-aware system operation. Our prediction framework consists of two stages: 1) using only the Slurm accounting logs as training data and 2) augmenting the training data with historical GPU profiling metrics collected with DCGM. The maximum GPU utilization predictions using only the Slurm submission features achieve up to 97% accuracy. Furthermore, features engineered from GPU-compute and memory activity metrics exhibit good correlations with average power utilization, and our runtime power usage prediction experiments result in up to 92% prediction accuracy. These findings demonstrate the effectiveness of DCGM metrics in capturing application characteristics and highlight their potential for developing predictive models to support dynamic power management in HPC systems.

GPU nodes are central to modern HPC and AI workloads, yet many failures do not manifest as immediate hard faults. While some instabilities emerge gradually as weak thermal or efficiency drift, a significant class occurs abruptly with little or no numeric precursor. In these detachment-class failures, GPUs become unavailable at the driver or interconnect level and the dominant observable signal is structural, including disappearance of device metrics and degradation of monitoring payload integrity. This paper proposes an observability-aware early-warning framework that jointly models (i) utilization-aware thermal drift signatures in GPU telemetry and (ii) monitoring-pipeline degradation indicators such as scrape latency increase, sample loss, time-series gaps, and device-metric disappearance. The framework is evaluated on production telemetry from GPU nodes at GWDG, where GPU, node, monitoring, and scheduler signals can be correlated. Results show that detachment failures exhibit minimal numeric precursor and are primarily observable through structural telemetry collapse, while joint modeling increases early-warning lead time compared to GPU-only detection. The dataset used in this study is publicly available at https://doi.org/10.5281/zenodo.19052367.

Deep learning training commonly runs on shared multi-tenant GPU servers, where exclusive allocation provides isolation but can leave resources underutilized and increase queueing time. Collocation can improve efficiency, but interference-agnostic placement may cause severe slowdowns, while inaccurate memory information can lead to out-of-memory (OOM) failures. We present AEGIS, a server-scale runtime scheduling system for controlled collocation of deep learning training workloads on shared multi-GPU servers. AEGIS integrates memory feasibility, post-placement observation, runtime-pressure filtering, placement, and OOM-aware recovery in a single scheduling loop. After placement, AEGIS observes workload activity before permitting further collocation, then uses low-overhead telemetry to determine whether a GPU can safely accept additional work. OOM failures trigger retries under progressively safer memory conditions, eventually falling back to exclusive execution. This online approach avoids costly offline pairwise compatibility profiling. We evaluate AEGIS using vision, Transformer, recommendation, and LLM-style workloads across three production-derived traces. AEGIS reduces geometric-mean makespan by 16% relative to Lucid, 21% relative to Horus, and 27% relative to exclusive allocation. Sensitivity studies show that activity-anchored observation and runtime-pressure filtering balance conservative isolation against interference-agnostic collocation, improving makespan while limiting sharing-induced per-task slowdown.