GPU Monitoring
This feature is introduced in version 10.3.
The GPU Monitoring feature enhances the application optimization process by adding GPU usage metrics to the existing criteria for monitoring CPU utilization, I/O operations, keyboard input, and mouse interactions.
GPU activity is treated as a critical indicator when assessing application inactivity. This is especially important for modern workloads such as simulations and AI applications, which heavily depend on GPU resources. In these cases, GPU utilization is a more accurate indicator of true inactivity, as traditional indicators—such as I/O operations, keyboard input, and mouse interactions—can be misleading.
GPU Monitoring collects application-level GPU utilization, expressed as a percentage. This metric reflects the average usage of all GPU engines by an application process during a specified sampling interval.
The collected GPU utilization metric is integrated into the application optimization process to evaluate application activity and inactivity. It complements existing optimization metrics, including CPU utilization, I/O operations, keyboard strokes, and mouse interactions.
No additional GPU metrics are collected. GPU Monitoring does not collect information on GPU memory usage, temperature, power consumption, or detailed per-engine metrics.
To enable GPU utilization tracking, update the cold_master.conf configuration file and set the gpu attribute in the application optimization settings.
GPU Monitoring is available only on Windows systems and supports NVIDIA, AMD, and Intel GPUs.
This feature has been tested on the following hardware: NVIDIA GeForce MX330, AMD Radeon Vega Mobile Gfx, and Intel® Iris® Xe Graphics.