CudaMon: GPU Monitoring for R

Extended System and GPU Monitoring with Collectl and NVML

Quick Start

library(CudaMon)

# Start monitoring with GPU tracking
proc <- cl_start("my_workflow", monitor_gpu = TRUE, gpu_monitor_type = "nvml")

# Your GPU computation here
cl_timestamp(proc, "computation_start")
# ... your code ...
cl_timestamp(proc, "computation_end")

# Stop and visualize
cl_stop(proc)
cl_plot_system_metrics(proc)  # 4-panel CPU/GPU plot

Features

• System Monitoring: CPU, memory, disk I/O via collectl
• GPU Monitoring: Two backends:
  • collectl -sG (basic)
  • NVML (NVIDIA GPUs - detailed metrics)
• Background Collection: Continuous GPU metrics during computation
• Integrated Plots: Combined CPU/GPU usage visualization

Installation

devtools::install_github("aminzadenoori/CudaMon")

System Requirements: collectl utility, NVIDIA drivers (for NVML)

API

• cl_start() / cl_stop() - Start/stop monitoring
• cl_timestamp() - Mark workflow phases
• cl_get_gpu_metrics() - Access real-time GPU data
• cl_plot_system_metrics() - Generate usage plots

Example Output

Shows CPU usage, memory, GPU utilization, and GPU memory over time.

---

Extends Rcollectl with GPU monitoring capabilities for profiling accelerated computations.