IMX93 NPU

James33 — Mon, 17 Nov 2025 09:48:19 GMT

Dear NXP Community Experts/Engineers,
Hello! Thank you for taking the time to review my question.

I'm trying to use NPU to accelerate inference on the IMX9352 development board.
-chip: imx9352
-yocto version: scarthgap 6.6.52

I read the document <i.MX Machine Learning User's Guide.pdf>, followed the steps in Section 6.2.3, and the console output incorrect print information.
commands:
# cd /usr/bin/ethosu/examples
# cp ../../tensorflow-lite-2.16.2/examples/labels.txt ./
# cp ../../tensorflow-lite-2.16.2/examples/grace_hopper.bmp ./

# vela ../../tensorflow-lite-2.16.2/examples/mobilenet_v1_1.0_224_quant.tflite

# ./inference_runner -n ./output/mobilenet_v1_1.0_224_quant_vela.tflite -i grace_hopper.bmp -l labels.txt -o output.txt

error message:
root@imx93-11x11-lpddr4x-evk:/usr/bin/ethosu/examples# ./inference_runner -n ./output/mobilenet_v1_1.0_224_quant_vela.tflite -i grace_hopper.bmp -l labels.txt -o output.txt
Send Ping
Send version request
Send capabilities request
Capabilities:
version_status:1
version:{ major=0, minor=0, patch=0 }
product:{ major=6, minor=0, patch=0 }
architecture:{ major=1, minor=0, patch=6 }
driver:{ major=0, minor=16, patch=0 }
macs_per_cc:8
cmd_stream_version:0
custom_dma:false
Create network
Error: IOCTL failed

According to Section 8.1.2, the object detection case can run normally, but if the parameter "-d /usr/lib/libethosu_delegate.so" is added, it cannot run.
here is the logs:

root@imx93-11x11-lpddr4x-evk:/usr/bin/eiq-examples-git/object_detection# python3 main.py -i /dev/video1 -d /usr/lib/libethosu_delegate.so
[ WARN:0@1.827] global cap_gstreamer.cpp:2839 handleMessage OpenCV | GStreamer warning: Embedded video playback halted; module source reported: Could not read from resource.
[ WARN:0@1.831] global cap_gstreamer.cpp:1698 open OpenCV | GStreamer warning: unable to start pipeline
[ WARN:0@1.831] global cap_gstreamer.cpp:1173 isPipelinePlaying OpenCV | GStreamer warning: GStreamer: pipeline have not been created
INFO: Ethosu delegate: device_name set to /dev/ethosu0.
INFO: Ethosu delegate: cache_file_path set to .
INFO: Ethosu delegate: timeout set to 60000000000.
INFO: Ethosu delegate: enable_cycle_counter set to 0.
INFO: Ethosu delegate: enable_profiling set to 0.
INFO: Ethosu delegate: profiling_buffer_size set to 2048.
INFO: Ethosu delegate: pmu_event0 set to 0.
INFO: Ethosu delegate: pmu_event1 set to 0.
INFO: Ethosu delegate: pmu_event2 set to 0.
INFO: Ethosu delegate: pmu_event3 set to 0.
INFO: EthosuDelegate: 108 nodes delegated out of 111 nodes with 1 partitions.
INFO: Created TensorFlow Lite XNNPACK delegate for CPU.
Traceback (most recent call last):
File "/usr/bin/eiq-examples-git/object_detection/main.py", line 68, in <module>
interpreter.invoke()
File "/usr/lib/python3.12/site-packages/tflite_runtime/interpreter.py", line 941, in invoke
self._interpreter.Invoke()
RuntimeError: Ethos_u inference failed
Node number 111 (EthosuDelegate) failed to invoke.

I want to know how to test whether my NPU can work properly and how to use it correctly for accelerated inference. I have an object detection model, detect.tflite(SSD_V2), and I hope to use the NPU on IMX93 to accelerate the inference function of this model.

The complete print information can be found in the attachment "logs.txt".

Thank you again for your assistance! Please let me know if you need any additional information.I will provide it promptly.

Best regards,

James

Re: IMX93 NPU

danielchen — Wed, 19 Nov 2025 03:01:58 GMT

Hi @James33

I tried the getting started guide on my side , it runs very well. I am using LF6.12.34 (the latest version). I attached the logs here. I would suggest you try this version again.

root@imx93evk:/usr/bin/ethosu/examples# ./inference_runner -n ./output/mobilenet_v1_1.0_224_quant_vela.tflite .txt
Send Ping
Send version request
Send capabilities request
Capabilities:
version_status:1
version:{ major=0, minor=0, patch=0 }
product:{ major=6, minor=0, patch=0 }
architecture:{ major=1, minor=0, patch=6 }
driver:{ major=0, minor=16, patch=0 }
macs_per_cc:8
cmd_stream_version:0
custom_dma:false
Create network
Create inference
output offset(150528)..size(1001)....
Wait for inference
Inference status: ok
Output size: 1001

Detected: military uniform, confidence:72
root@imx93evk:/usr/bin/ethosu/examples#
root@imx93evk:/usr/bin/ethosu/examples#
root@imx93evk:/usr/bin/ethosu/examples#
root@imx93evk:/usr/bin/ethosu/examples#
root@imx93evk:/usr/bin/ethosu/examples#
root@imx93evk:/usr/bin/ethosu/examples# cd /usr/bin/tensorflow-lite-2.19.0/examples/
root@imx93evk:/usr/bin/tensorflow-lite-2.19.0/examples# ./label_image -m ../../ethosu/examples/output/mobilenelegate_path=/usr/lib/libethosu_delegate.so
INFO: Loaded model ../../ethosu/examples/output/mobilenet_v1_1.0_224_quant_vela.tflite
INFO: resolved reporter
INFO: Ethosu delegate: device_name set to /dev/ethosu0.
INFO: Ethosu delegate: cache_file_path set to .
INFO: Ethosu delegate: timeout set to 60000000000.
INFO: Ethosu delegate: enable_cycle_counter set to 0.
INFO: Ethosu delegate: enable_profiling set to 0.
INFO: Ethosu delegate: profiling_buffer_size set to 2048.
INFO: Ethosu delegate: pmu_event0 set to 0.
INFO: Ethosu delegate: pmu_event1 set to 0.
INFO: Ethosu delegate: pmu_event2 set to 0.
INFO: Ethosu delegate: pmu_event3 set to 0.
INFO: EXTERNAL delegate created.
INFO: EthosuDelegate: 1 nodes delegated out of 1 nodes with 1 partitions.
INFO: Applied EXTERNAL delegate.
INFO: invoked
INFO: average time: 3.759 ms
INFO: 0.780392: 653 military uniform
INFO: 0.105882: 907 Windsor tie
INFO: 0.0156863: 458 bow tie
INFO: 0.0117647: 466 bulletproof vest
INFO: 0.00784314: 835 suit

Regards

Daniel

topic Re: IMX93 NPU in i.MX Processors

IMX93 NPU

Re: IMX93 NPU