LLM Infrastructure Telemetry Dataset | NVIDIA P100 + Dual Xeon | 10K+ Labeled Rows

First-party, non-synthetic operational telemetry collected directly from Scriptores Helm LLC's Dell dual-Xeon LLM host and NVIDIA Tesla P100 PCIe 16 GB GPU during normal operation. One-second rows combine NVML sensor measurements with aggregate Linux host counters. Inputs include GPU utilization, memory use, temperature, board power, power limit, P-state, clocks, CPU utilization, load averages, RAM and swap, storage I/O, and network byte deltas. Seven ground-truth columns record GPU temperature, GPU board power, host CPU utilization, one-minute load, RAM use, disk writes, and received network traffic directly measured at least 60 seconds after each feature row. These fields support regression, forecasting, anomaly detection, and systems-operations evaluation. No workload is generated for the dataset and no values are synthetic or AI-generated. Privacy and exclusions: The collector does not inspect or store prompts, model responses, files, process names, command lines, usernames, IP addresses, UUIDs, customer data, or personal data. It records hardware sensors and aggregate host counters only. Delivery: versioned CSV with 10,000+ labeled rows and 31 columns, a 500-row sample, full column dictionary, collection methodology, validation results, limitations, and SHA-256 checksums. The collection represents one P100 and dual-Xeon machine and one October 2026 time window; it is not presented as a benchmark for all hardware or workloads. Nonexclusive licensing for internal AI traini