This message provides a number indicating the total time in milliseconds the specific device was used for acceleration.
The threshold number is 0 ms. This helps identify if a device was actually used during the run of the application. If multiple devices are available to accelerate the implementation, consider additional load balancing between the devices.
If the workload to be accelerated is sufficiently large and the algorithm supports parallel processing of the task, this indicates that not all available compute resources are sufficiently used. Consider running additional kernel(s) or sections of the overall algorithm on the remaining devices.