CUDA doesnt work

Hello,

I am currently working with the ZED Box Orin 16GB. While there are no issues detected in the ZED_Diagnostics, running nvidia-smi fails to detect the GPU. It seems that CUDA loading has failed.

I would appreciate your support in resolving this issue. The installed SDK version is L4T 36.3, and although the sample code runs, it executes very slowly.

Thank you in advance for your assistance!
},
“Processor”: {
“L4T version”: “36.3.0”,
“OS”: "Ubuntu 22.04.5 LTS ",
“avxSupported”: false,
“coreCount”: “8”,
“cpu”: “”,
“datetime”: “2025-01-17 08:31:33 AM”,
“hyperThreading”: “0”,
“motherboard”: “Jetson, NVIDIA”,
“ramCapacity”: “16.416158”,
“ramUsage”: “”,
“ramUsed”: “5.094449”,
“threadCount”: “8”,
“valid”: [
"Processor: “,
Motherboard: Jetson, NVIDIA”
]
},
“ZED SDK”: {
“CUDA Toolkit version”: “V12.2.140 -->located in /usr/local/cuda-12.2/bin/nvcc”,
“CUDA loading”: false,
“GPU Driver Version”: “Driver Version : N/A”,
“GPU Performance”: " Performance State : N/A”,
“ZED SDK Version (Diag)”: “4.2.3”,
“ZED SDK Version (RT)”: “4.2.3”,
“ai”: {
“checkAI”: true,
“cudaVersion”: 12020,
“cudnnCudartVersion”: 12020,
“cudnnVersion”: 8904,
“cudnnVersionExpected”: 8904,
“tensorVersion”: 8602
},
“binFiles”: [
“libsl_ai.so”,
“libsl_zed.so”,
“libsl_zed_static.a”
],
“resourcesFiles”: [
“neural_depth_3.6.model”,
“objects_accurate_3.2.model”,
“objects_medium_3.2.model”,
“objects_performance_3.2.model”,
“person_head_accurate_2.6.model”,
“person_head_performance_2.6.model”,
“person_reid_1.4.model”,
“skeleton_body18_3.2.model”,
“skeleton_body38_3.5.model”
],
“resourcesFilesExpected”: [
“objects_performance_3.2”
],
“valid”: [
ZED SDK version: 4.2.3”,
CUDA version: V12.2.140”
]
}
}

Hi @Hamin242
Welcome to the Stereolabs community.

nvidia-smi does not work on Jetson.

Please define very slowly. What FPS? What depth mode are you using?
Is ZED_Depth_Viewer working correctly? What FPS can you read at the bottom left of the GUI?

Thank you for the quick response. Currently, when using the zed depth viewer, it runs well but 5~10 FPS and the command line shows the message “too slow.” In the custom object detection, the inference time exceeds 100ms.



AI Models
valid
0 MULTI CLASS DETECTION is optimized
1 MULTI CLASS MEDIUM DETECTION is optimized
2 MULTI CLASS ACCURATE DETECTION is optimized
3 HUMAN BODY FAST DETECTION is optimized
4 HUMAN BODY MEDIUM DETECTION is optimized
5 HUMAN BODY ACCURATE DETECTION is optimized
6 HUMAN BODY 38 FAST DETECTION is optimized
7 HUMAN BODY 38 MEDIUM DETECTION is optimized
8 HUMAN BODY 38 ACCURATE DETECTION is optimized
9 PERSON HEAD DETECTION is optimized
10 PERSON HEAD ACCURATE DETECTION is optimized
11 REID ASSOCIATION is optimized
12 NEURAL DEPTH is optimized
13 NEURAL PLUS DEPTH is optimized
Camera Test
InternalDevicesCount 0
ZEDCount 1
camera
Device ID 0
Firmware 2001
Init Output OK
Model ZED-X
Serial Number 42461326
valid
0 Camera: ZED-X
1 Firmware: 2001
resolutions
HD1200@30
fps (input) 30
fps (output) 30
height (input) 1200
height (output) 1200
initialization OK
status true
width (input) 1920
width (output) 1920
HD1200@60
fps (input) 60
fps (output) 60
height (input) 1200
height (output) 1200
initialization OK
status true
width (input) 1920
width (output) 1920
SVGA@120
fps (input) 120
fps (output) 120
height (input) 600
height (output) 600
initialization OK
status true
width (input) 960
width (output) 960
sensors {}
Devices
CorruptedFirmware false
GMSL driver ii stereolabs-zedbox-onx16 1.2.1-LI-MAX96712-all-ZEDBOX-L4T36.3.0 arm64 NVIDIA Kernel DTB Modified Package by Stereolabs\n
GMSL driver compatiblity OK
GMSLList
0
Serial 42461326
Sub Model ZED-XG 120mm
id 0
path /dev/i2c-9
state Available
USBList
0
USB_path /3
idProduct 0xc52b
idVendor 0x046d
1
USB_path /2/4/3
idProduct 0x6015
idVendor 0x0403
2
USB_path /2/4/2
idProduct 0x6015
idVendor 0x0403
3
USB_path /2/4/1
idProduct 0x01a9
idVendor 0x1546
4
USB_path /2/4
idProduct 0x0608
idVendor 0x05e3
5
USB_path /2/3
idProduct 0x0032
idVendor 0x8087
6
USB_path /2
idProduct 0x0608
idVendor 0x05e3
Graphics Card
deviceCount 1
deviceDriverVersion 12020
devices
0
arch Ampere
computeCapability 8.7
cores 512
name Orin
totalMemoryBytes 16416161792
totalMemoryMB 15655.671875
glx_info OpenGL vendor string: NVIDIA Corporation\nOpenGL renderer string: NVIDIA Tegra Orin (nvgpu)/integrated\nOpenGL core profile version string: 4.6.0 NVIDIA 540.3.0\nOpenGL core profile shading language version string: 4.60 NVIDIA\nOpenGL core profile context flags: (none)\nOpenGL core profile profile mask: core profile\nOpenGL core profile extensions:\nOpenGL version string: 4.6.0 NVIDIA 540.3.0\nOpenGL shading language version string: 4.60 NVIDIA\nOpenGL context flags: (none)\nOpenGL profile mask: (none)\nOpenGL extensions:\nOpenGL ES profile version string: OpenGL ES 3.2 NVIDIA 540.3.0\nOpenGL ES profile shading language version string: OpenGL ES GLSL ES 3.20\nOpenGL ES profile extensions:\n
initResult 0
valid
0 Graphics card: Orin
Processor
L4T version 36.3.0
OS Ubuntu 22.04.5 LTS
avxSupported false
coreCount 8
cpu
datetime 2025-01-17 06:31:00 AM
hyperThreading 0
motherboard Jetson, NVIDIA
ramCapacity 16.416162
ramUsage
ramUsed 8.619586
threadCount 8
valid
0 Processor:
1 Motherboard: Jetson, NVIDIA
ZED SDK
CUDA Toolkit version V12.2.140 -->located in /usr/local/cuda-12.2/bin/nvcc
CUDA loading false
GPU Driver Version Driver Version : N/A
GPU Performance Performance State : N/A
ZED SDK Version (Diag) 4.2.3
ZED SDK Version (RT) 4.2.3
ai
checkAI true
cudaVersion 12020
cudnnCudartVersion 12020
cudnnVersion 8904
cudnnVersionExpected 8904
tensorVersion 8602
binFiles
0 libsl_ai.so
1 libsl_zed.so
2 libsl_zed_static.a
resourcesFiles
0 neural_depth_3.6.model
1 objects_accurate_3.2.model
2 objects_medium_3.2.model
3 objects_performance_3.2.model
4 person_head_accurate_2.6.model
5 person_head_performance_2.6.model
6 person_reid_1.4.model
7 skeleton_body18_3.2.model
8 skeleton_body38_3.5.model
resourcesFilesExpected
0 objects_performance_3.2
valid
0 ZED SDK version: 4.2.3
1 CUDA version: V12.2.140

What NVIDIA Jetson Orin model are you using? Nano, NX, AGX?
What power mode?
Is the jetson_clocks script running?
What camera grab resolution?
What Depth mode and Object Detection model?

  1. using nano model
  2. 15W
  3. jetson_clocks --show is running
  4. HD1200
  5. depth mode: ULTRA / object detection : yolv8s

How can check GPU using like nvidia-smi ? Is other tools or command ?

Only download below file. If restalling, can load CUDA?

ZED SDK for JetPack 6.0 GA (L4T 36.3) 4.2 (Jetson Orin, CUDA 12.2)

Hi @Hamin242

Jetson Orin Nano or Jetson Nano?

Try lowering the resolution to reduce the computing requirement.

Is your YOLO model optimized?

The issue doesn’t seem to be related to the model or resolution but rather that GPU-accelerated processing doesn’t appear to be working. In the ZED Diagnostics, the CUDA_loading field is showing as False. Could you provide a solution for this? Or is parallel processing still being performed even if it shows as False?

This is not a problem. This field is relevant only when using our SDK on desktop/laptop PCs with discrete NVIDIA GPU.
The Jetson devices have different behaviors concerning how CUDA is used.

Please note that “without CUDA”, the ZED SDK cannot work correctly and it cannot generate depth information.
If CUDA is not available you surely receive many warnings.

We can then continue from my last question:

I’m using the an ultralytics YOLOv8n model.

What model size is appropriate for real-time object detection? I’m planning to customize the YOLO-NAS model - is this possible with a quantized model?

Additionally, my WiFi adapter is not being detected. It was previously working, but after a reboot it’s now showing as wifi(E) with an error. Is flashing the only reset method available?

Hi @Hamin242
have you tried to use this example?

Hi @Myzhar
Thank you for your response. I am currently working on object detection using Python.(zed-sdk/object detection/custom detector/python/pytorch_yolov8 at master · stereolabs/zed-sdk · GitHub)

At the moment, I need Wi-Fi to retrieve the sample, but the Wi-Fi adapter is not being recognized. I ran the command dmesg | grep iwlwifi and discovered an iwlwifi(E) error indicating a module verification failure (module verification failed).

Could you please let me know if there is an alternative solution to resolve this issue without performing a reset, instead of the method described at Reset your ZED Box Orin NX - Stereolabs?

Hi @Hamin242
what did you change to break the WiFi communication?
Can you please send me the output of the commands sudo lspci and cat /etc/nv_tegra_release?

Hi @Myzhar
5days ago, I received a notification that PC automatically updated the Ubuntu OS, and since then it seems that the Wi-Fi is not working.


Hi @Hamin242
thank you. This information can be relevant to understand what’s happened.

For connect Wi-Fi, have to reset ZED BOX? Reset your ZED Box Orin NX - Stereolabs

Hi @Hamin242
a reset will allow you to use it again, but if the problem comes from a system update you could face it again later.

The hardware team is investigating the problem searching for the cause.
We will provide you wit the instructions to install the driver soon.

@Hamin242 an additional question: did you perform a distribution upgrade to 24.04 or a simple software package update and you are still on Ubuntu 22.04?

Hi, still Ubuntu 22.04.5LTS