Resource efficiency and potential savings metrics for AI workloads
The Efficiency and Potential Savings metrics dashboard in Cloudera Observability provides detailed insights into the efficiency of your AI workloads, including usage percentages and potential savings.
- CPU: Displays the percentage of CPU resources utilized by the AI
workload job execution.
Potential savings are displayed by multiplying
the CPU cost defined at chargeback setup * unused CPU core hours. - Memory: Indicates the total memory consumed by the AI workload
job, represented as a percentage. This is calculated using the peak memory used during
execution, measured in gigabytes multiplied by milliseconds.
Potential savings are displayed by multiplying
the memory cost defined at chargeback setup * unused Memory GB hours. - Overall: Displays the average usage for both CPU, GPU, memory,
and GPU memory. The percentage is calculated as
(CPU percentage + Memory percentage) / 2.Potential savings are calculated by adding potential savings from all resources:
(CPU Potential Savings + Memory Potential Savings).
These metrics help you identify the under-utilization of resources. High CPU or memory wastage may suggest the need to reallocate resources, optimize usage, or adjust configurations to allocate fewer resources.
Dashboard color indicators:
- Green: Usage is between 75% and 100%—indicating efficient utilization.
- Orange: Usage is between 25% and 74%—indicating moderate utilization.
- Blue: Usage is between 0% and 24%—indicating low utilization.
