CodeProject.AI Discussions

Re: Codeproject AI Object detection

24-Nov-23 5:12

Before installing Codeproject.AI did you install CUDA 11.8 and cuDNN for CUDA 11.x

Member 852239724-Nov-23 5:17

Member 8522397

24-Nov-23 5:17

Hi Mike
I'll try
Thanks

Re: Codeproject AI Object detection

Member 1600435725-Nov-23 18:11

Member 16004357

25-Nov-23 18:11

I found the same thing, it says CUDA drivers are there, then says it's not present.

CUDA not available since version 2.3.4

StefanLU24-Nov-23 2:23

24-Nov-23 2:23

I didn't update any graphics drivers and it worked fine until version 2.2.4.

Now the installation of version 2.3.4 claims that CUDA isn't available and GPU mode can't be activated any longer.

Any ideas? Unfortunately I must restore the last version now, but would appreciate if I could update to the newest one.

Mike Lud24-Nov-23 3:13

24-Nov-23 3:13

After updating to 2.3.4 did you update the modules?
Also can you provide more details, which module and install log for the module.

StefanLU24-Nov-23 5:57

24-Nov-23 5:57

No I didn't update the modules. But I tried the newest NVidia driver which changed the CUDA from 12.0 to 12.2.

That was the log during installation which showed, that it didn't recognize the installed CUDA:

Reinstalling 2.2.4 this part looked like this, without any other change (new NVidia driver still installed):

That's now the "old" reinstalled 2.2.4. I always use only "Face Processing" (currently disabled) and "YOLO v5.6.2". License plate doesn't work, it never finds anything so I stopped testing it.

Currently I've only the LOG for 2.2.4 which works. But the difference of 2.3.4 is only that it claims CUDA present = false:

Installation Log
2023-11-24 12:46:30:              Installing CodeProject.AI Analysis Module                
2023-11-24 12:46:30: ========================================================================
2023-11-24 12:46:30:                    CodeProject.AI Installer                             
2023-11-24 12:46:30: ========================================================================
2023-11-24 12:46:30: Checking GPU support                                                  
2023-11-24 12:46:30: CUDA Present...True
2023-11-24 12:46:30: Allowing GPU Support: Yes
2023-11-24 12:46:30: Allowing CUDA Support: Yes
2023-11-24 12:46:30: General CodeProject.AI setup                                          
2023-11-24 12:46:30: Creating Directories...Done
2023-11-24 12:46:30: Processing Core SDK                                                   
2023-11-24 12:46:30: Installing module ObjectDetectionYolo 1.6.1                           
2023-11-24 12:46:30: Checking for python37 download...Present
2023-11-24 12:46:30: Creating Virtual Environment (Shared)...Python 3.7 Already present
2023-11-24 12:46:30: Enabling our Virtual Environment...Done
2023-11-24 12:46:30: Confirming we have Python 3.7...present
2023-11-24 12:46:32: Ensuring Python package manager (pip) is installed...Done
2023-11-24 12:46:32: Ensuring Python package manager (pip) is up to date...Done
2023-11-24 12:46:32: Choosing Python packages from requirements.windows.cuda.txt
2023-11-24 12:49:55: Installing Packages into Virtual Environment...Success
2023-11-24 12:50:05: Downloading Standard YOLO models...Expanding...Tar failed - moving to PowerShell...Done.
2023-11-24 12:50:18: Downloading Custom YOLO models...Expanding...Tar failed - moving to PowerShell...Done.
2023-11-24 12:50:18: Installing Server SDK support:
2023-11-24 12:50:20: Ensuring Python package manager (pip) is installed...Done
2023-11-24 12:50:21: Ensuring Python package manager (pip) is up to date...Done
2023-11-24 12:50:21: Choosing Python packages from requirements.txt
2023-11-24 12:50:27: ERROR: pip's dependency resolver does not currently take into account all the packages that are installed. This behaviour is the source of the following dependency conflicts.
2023-11-24 12:50:27: botocore 1.32.6 requires urllib3<1.27,>=1.25.4; python_version < "3.10", but you have urllib3 2.0.7 which is incompatible.
2023-11-24 12:50:27: Installing Packages into Virtual Environment...Success
2023-11-24 12:50:27: Setup complete                                                        
Installer exited with code 0

modified 24-Nov-23 13:53pm.

StefanLU25-Nov-23 23:36

25-Nov-23 23:36

Any idea how to solve this? If necessary I repeat the installation and safe the log but as I already mentioned the only difference is the detection result for CUDA 2.3.4 -> false and 2.2.4 -> true. All other settings and drivers are unchanged.

Member 1614350026-Nov-23 5:00

Member 16143500

26-Nov-23 5:00

Maybe try CUDA 11.7? It's working for me with an RTX 3060 and AIServer 2.3.4.

Chris Maunder27-Nov-23 10:17

Chris Maunder

27-Nov-23 10:17

At a terminal window, can you run the nvidia-smi command and then the nvcc --version command and let me know what it shows?

cheers
Chris Maunder

StefanLU27-Nov-23 14:07

27-Nov-23 14:07

"nvcc --version" is unknown

Mike Lud27-Nov-23 14:27

27-Nov-23 14:27

Since nvcc --version did not work it looks like you do not have CUDA installed.
Post a screenshot of Windows Apps & features searching CUDA

StefanLU28-Nov-23 8:12

28-Nov-23 8:12

Very interesting. You are right. But I wonder why it works without any issue with version 2.2.4 and older. I thought, all parts were installed with the NVidia driver.

and

I'm downloading the current version and try it again.

Jeepin 9529-Nov-23 5:26

Jeepin 95

29-Nov-23 5:26

I am having the same issue upgrading from 2.1 to 2.3.4. GPU was working great with 2.1 and now I am unable to see GPU as an option. I have a GTX 1050 in this Windows 11 system.

After seeing that GPU wasn't there for 2.3.4 I updated the NVIDIA drivers to the current version (546.17). Finding this thread I also downloaded and installed NVIDIA CUDA Toolkit 12.3 which forced a reboot and I still don't see GPU as an option. Using the Install Modules tab I uninstalled Face Processing and all of the Object Detection modules, then re-installed YOLOv5 6.2. In the logs I also see "07:13:56:ObjectDetectionYolo: CUDA Present...No".

nvidia-smi gives me this:

+---------------------------------------------------------------------------------------+
| NVIDIA-SMI 546.17                 Driver Version: 546.17       CUDA Version: 12.3     |
|-----------------------------------------+----------------------+----------------------+
| GPU  Name                     TCC/WDDM  | Bus-Id        Disp.A | Volatile Uncorr. ECC |
| Fan  Temp   Perf          Pwr:Usage/Cap |         Memory-Usage | GPU-Util  Compute M. |
|                                         |                      |               MIG M. |
|=========================================+======================+======================|
|   0  NVIDIA GeForce GTX 1050      WDDM  | 00000000:01:00.0 Off |                  N/A |
| N/A   45C    P8              N/A / ERR! |      0MiB /  3072MiB |      0%      Default |
|                                         |                      |                  N/A |
+-----------------------------------------+----------------------+----------------------+

+---------------------------------------------------------------------------------------+
| Processes:                                                                            |
|  GPU   GI   CI        PID   Type   Process name                            GPU Memory |
|        ID   ID                                                             Usage      |
|=======================================================================================|
|  No running processes found                                                           |
+---------------------------------------------------------------------------------------+

nvcc command isn't found and NVIDA CUDA Toolkit 12.3 is installed.

Mike Lud29-Nov-23 7:28

29-Nov-23 7:28

Try rebooting then run nvcc - version

Jeepin 9529-Nov-23 8:42

Jeepin 95

29-Nov-23 8:42

I was initially forced to reboot by the CUDA installation and have rebooted twice since, nvcc is still not there. This is on my BlueIris machine so I do not have Visual Studio installed. When installing CUDA it did give a warning that some things may not work due to not having Visual Studio, but it doesn't seem like that should be a requirement for running the AI server.

After a complete uninstall of 2.3.4 and re-installing 2.1.11 (the latest version I had working) during the YOLO module installation I do get

10:31:59:ObjectDetectionYolo: CUDA Present...True

in the logs and GPU works fine.

StefanLU29-Nov-23 22:15

29-Nov-23 22:15

That's same for me. The issue appears after reinstalling (I always remove the old version) with version 2.3.4. Downgrading (same way, complete reinstall) to 2.2.4 as my last stable version it detects CUDA and works.

It seems that there was a change in detection of CUDA version between 2.2.4 and 2.3.4. Later I retry the installation of 2.3.4. Hopefully it works, because I don't want to freeze version 2.2.4 as last running version, especially because BI may cause issues with newer versions thereafter.

Jeepin 9530-Nov-23 5:08

Jeepin 95

30-Nov-23 5:08

I was now able to get 2.3.4 working with CUDA on my GTX 1050 machine. Since the CUDA install warned me that Visual Studio wasn't installed and nvcc wasn't found I decided to go back and install Visual Studio Community first. I then re-installed CUDA 12.3 which no longer warned me about missing VS and reported some addition items installed when it completed. Running nvcc from the command line was now found and reported some details.

I then uninstalled and deleted the remaining files from my CodeProject.AI install, rebooted, then re-installed the server. CUDA Present...Yes (version 12.3) was reported in the status logs this time and Face and Object Detection are now able to use the GPU.

So something certainly changed between versions and there is some requirement missing.

Chris Maunder30-Nov-23 3:05

Chris Maunder

30-Nov-23 3:05

I've pinned down the reason and it's an issue where we're assuming nvcc will always be available in Windows when CUDA is installed, which isn't the case. This issue doesn't affect Linux (we check nvcc and nvidia-smi).

For those who can't wait:

In C:\Program Files\CodeProject\AI\SDK\Scripts is a file utils.bat. In that is a method :GetCudaVersion. Replace what you have with this:

CMD

:GetCudaVersion

    :: Use nvcc to find the CUDA version
    where nvcc >nul 2>&1
    if !errorlevel! == 0 (
        REM Get the line containing "release x.y"
        for /f "tokens=*" %%i in ('nvcc --version ^| findstr /C:"release"') do set cudaLine=%%i
        REM Get the 5th token in the line when split by , and spaces
        for /f "tokens=5 delims=, " %%a in ("!cudaLine!") do set cuda_version=%%a
    ) else (
        REM Backup attempt: Use nvidia-smi to find the CUDA version
        where nvidia-smi >nul 2>&1
        if !errorlevel! == 0 (
            REM Get the line containing "CUDA Version x.y"
            for /f "tokens=*" %%i in ('nvidia-smi ^| findstr /C:"CUDA Version"') do set cudaLine=%%i
            REM Get the 9th token in the line when split by spaces
            for /f "tokens=9 delims= " %%a in ("!cudaLine!") do set cuda_version=%%a
        ) else (
            REM echo Unable to find nvcc or nvidia-smi
        )
    )

    if "!cuda_version!" neq "" (
        for /f "tokens=1,2 delims=." %%a in ("!cuda_version!") do (
            set "cuda_major_version=%%a"
            exit /b
        )
    )

    exit /b

cheers
Chris Maunder

modified 30-Nov-23 9:43am.

StefanLU30-Nov-23 4:28

30-Nov-23 4:28

Thank you for troubleshooting. I wait for the fix in next version.