Hello,
I am currently working on benchmarking AI models on an Intel® Core™ Ultra 7 265K processor.
My goal is to evaluate and compare performance across CPU, GPU, and NPU.
I have two main questions:
In my benchmarks, I observe that the CPU is often faster than the NPU for AI models.
For example, I am using the following command:
benchmark_app -m model.xml -d NPU -hint latency -t 15 -api async
Is this behavior expected? Under what conditions should the NPU outperform the CPU?
I would like to measure the power consumption of the NPU during inference. However, I was not able to find any official Intel tool that exposes this information.
I came across the following tool:
Could you confirm whether it is appropriate to use this tool for measuring NPU power consumption?
Additionally, is there any official reference or documentation providing typical minimum and maximum power consumption values for:
- CPU
- integrated GPU (iGPU)
- NPU
Thank you for your support.
