SC26 Artifact: GPU Workload Imbalance Telemetry Dataset

Pre-collected GPU telemetry dataset for SC26 2026 artifact evaluation. Contains 177,620 telemetry samples across 12 nodes x 4 NVIDIA A100 80GB GPUs (48 GPUs total), covering 30 imbalance scenarios with CV% range 0% to 173.21%. Approximately 14,800 rows per scenario per node. This dataset is the direct input to artifacts A2 (Standalone_Mode9.py) and A3 (Unified_Pipeline.py). Reviewers do not need to run A1 —download this dataset and run A2 and A3 directly (CPU-only, ~10-20 min). Folder structure after unzipping:SC26_data/ r04gn01/ ... r05gn06/ (12 node folders, 30 CSV files each) Cluster name and institutional details withheld for double-anonymous review. Hardware: NVIDIA A100 80GB GPUs, NVLink, CUDA 12.x. Dataset is publicly available. Full hardware and institutional details will be disclosed upon paper acceptance.