Changing the NPU power mode from the default "performance" to "turbo" doesn't change its actual capabiility to run faster.
Despite running the command sudo xrt-smi configure -d <BDF> --pmode turbo reports it being correctly executed and consequently the modality to supposedly having been changed, in reality npu performances after the above command aren't affected at all as also stated by the command sudo xrt-smi validate -d <BDF> reporting the npu being stuck into "performance" mode and also confirmed by running a simple test:
executing into "flm run qwen3.5:2b" the prompt "a website can be made in 10 steps":
PERFORMANCE MODE
Average decoding speed: 23.8301 tokens/s
Average prefill speed: 30.7483 tokens/s
TURBO MODE
Average decoding speed: 23.8648 tokens/s
Average prefill speed: 31.7367 tokens/s
Changing the NPU power mode from the default "performance" to "turbo" doesn't change its actual capabiility to run faster.
Despite running the command
sudo xrt-smi configure -d <BDF> --pmode turboreports it being correctly executed and consequently the modality to supposedly having been changed, in reality npu performances after the above command aren't affected at all as also stated by the commandsudo xrt-smi validate -d <BDF>reporting the npu being stuck into "performance" mode and also confirmed by running a simple test: