Getting error for NVIDIA CudNN with Matlab 2019b in Windows 10

Hi,

I have installed Cuda9.2 along with cudNN following the instruction given in NVIDIA site in Windows 10. I am going to use deep learning in MATLAB 2019b.

When I used ---> coder.checkGpuInstall('full')

I got the below error and messages:

Compatible GPU : PASSED

CUDA Environment : PASSED

Runtime : PASSED

cuFFT : PASSED

cuSOLVER : PASSED

cuBLAS : PASSED

cuDNN Environment : FAILED (Unable to find the 'NVIDIA_CUDNN' environment variable. Set 'NVIDIA_CUDNN' to point to the root directory of a NVIDIA cuDNN installation.)

TensorRT Environment : FAILED (Unable to find the 'NVIDIA_TENSORRT' environment variable. Set 'NVIDIA_TENSORRT' to point to the root directory of a TensorRT installation.)

Profiling Environment : PASSED

Basic Code Generation : FAILED (Test GPU code generation failed with the error 'emlc:compilationError'. View report for further information: View report)

ans =

struct with fields:

gpu: 1

cuda: 1

cudnn: 0

tensorrt: 0

basiccodegen: 0

basiccodeexec: 0

deepcodegen: 0

deepcodeexec: 0

tensorrtdatatype: 0

profiling: 1

Can anyone please help to resolve the issue?

With regards

[abdul kaleem siddiqi]

same issue

[Susama Bagchi]

need to set the environment. Better to contact the MATLAB support services. My issue is resolved for cuDNN.

[shuang zhou]

1. download and install cudnn https://developer.nvidia.com/cudnn
2. download and install tensorrt https://developer.nvidia.com/tensorrt
3. add cudnn and tensorrt to system environment variables list then you will get info below.
4. run coder.checkGpuInstall('full')

>> getenv('NVIDIA_CUDNN')

'C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v9.2'

>> getenv('NVIDIA_TENSORRT')

'C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v9.2\TensorRT-7.1.3.4'

struct with fields:

gpu: 1

cuda: 1

cudnn: 1

tensorrt: 1

basiccodegen: 1

basiccodeexec: 1

deepcodegen: 0

deepcodeexec: 0

tensorrtdatatype: 1

profiling: 1

[kang yang]

Excuse me, has your problem been solved, I also encountered the same problem

[ChrisLyu]

>> setenv('NVIDIA_CUDNN','C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v10.0')

>> getenv('NVIDIA_CUDNN')

ans =

'C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v10.0'

>> coder.checkGpuInstall('full')

Compatible GPU : PASSED

CUDA Environment : PASSED

Runtime : PASSED

cuFFT : PASSED

cuSOLVER : PASSED

cuBLAS : PASSED

cuDNN Environment : PASSED

TensorRT Environment : FAILED (Unable to find the 'NVIDIA_TENSORRT' environment variable. Set 'NVIDIA_TENSORRT' to point to the root directory of a TensorRT installation.)

Profiling Environment : PASSED

Basic Code Generation : FAILED (Test GPU code generation failed with the error 'emlc:compilationError'. View report for further information: View report)

ans =

다음 필드를 포함한 struct:

gpu: 1

cuda: 1

cudnn: 1

tensorrt: 0

basiccodegen: 0

basiccodeexec: 0

deepcodegen: 0

deepcodeexec: 0

tensorrtdatatype: 0

profiling: 1

[ChrisLyu]

https://developer.nvidia.com/rdp/cudnn-archive

[ChrisLyu]

Run Matlab as administrator.

Answer 1

[xingxingcui]

coder.checkGpuInstall('full')

Compatible GPU : PASSED

CUDA Environment : PASSED

Runtime : PASSED

cuFFT : PASSED

cuSOLVER : PASSED

cuBLAS : PASSED

cuDNN Environment : FAILED (Error generated while determining cuDNN library version 'getcuDNNVersion.cpp

C:\Users\Administrator\AppData\Local\Temp\tp21ea7936_7a49_41f9_ba3f_c12c8c253a5c\getcuDNNVersion.cpp: fatal error C1001: 编译器中发生内部错误。

(编译器文件“f:\dd\vctools\compiler\cxxfe\sl\p1\c\p0io.c”，第 2739 行)

要解决此问题，请尝试简化或更改上面所列位置附近的程序。

请选择 Visual C++

“帮助”菜单上的“技术支持”命令，或打开技术支持帮助文件来获得详细信息。

')

TensorRT Environment : FAILED (Unable to find the 'NVIDIA_TENSORRT' environment variable. Set 'NVIDIA_TENSORRT' to point to the root directory of a TensorRT installation.)

Profiling Environment : PASSED

Basic Code Generation : FAILED (Test GPU code generation failed with the error 'emlc:compilationError'. View report for further information: View report)

ans =

struct with fields:

gpu: 1

cuda: 1

cudnn: 0

tensorrt: 0

basiccodegen: 0

basiccodeexec: 0

deepcodegen: 0

deepcodeexec: 0

tensorrtdatatype: 0

profiling: 1

>> getenv('NVIDIA_CUDNN')

ans =

'C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v10.1'

Matlab R2020a , win10

how to solve this issue?

[Susama Bagchi]

Please contact MATLAB support services. First you need to confirm which version of CUDA you need to install for 2020a version. Then you need to set the environment path.

[ChrisLyu]

>> setenv('NVIDIA_CUDNN','C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v10.0')

>> getenv('NVIDIA_CUDNN')

ans =

'C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v10.0'

>> coder.checkGpuInstall('full')

Compatible GPU : PASSED

CUDA Environment : PASSED

Runtime : PASSED

cuFFT : PASSED

cuSOLVER : PASSED

cuBLAS : PASSED

cuDNN Environment : PASSED

TensorRT Environment : FAILED (Unable to find the 'NVIDIA_TENSORRT' environment variable. Set 'NVIDIA_TENSORRT' to point to the root directory of a TensorRT installation.)

Profiling Environment : PASSED

Basic Code Generation : FAILED (Test GPU code generation failed with the error 'emlc:compilationError'. View report for further information: View report)

ans =

다음 필드를 포함한 struct:

gpu: 1

cuda: 1

cudnn: 1

tensorrt: 0

basiccodegen: 0

basiccodeexec: 0

deepcodegen: 0

deepcodeexec: 0

tensorrtdatatype: 0

profiling: 1

[ChrisLyu]

https://developer.nvidia.com/rdp/cudnn-archive

[ChrisLyu]

You need Visual studio 2017 or earlier.

Run Matlab as administrator.

Answer 2

[yulei ji]

I have the same problem.

coder.checkGpuInstall()

Compatible GPU : PASSED

CUDA Environment : PASSED

Runtime : PASSED

cuFFT : PASSED

cuSOLVER : PASSED

cuBLAS : PASSED

cuDNN Environment : PASSED (Warning: Deep learning code generation has been tested with cuDNN v7.5. The provided cuDNN library v7.6 may not be fully compatible.)

Basic Code Generation : FAILED (Test GPU code generation failed with the error 'emlc:compilationError'. View report for further information: View report)

What shuold I do

[Susama Bagchi]

Please contact MATLAB support services.

[Lo jungle]

have you solve this probelm？

Answer 3

[Tuong]

This is how I do it on Matlab R2023b + Window 10 x64

First I install CUDA 11.8

Then I install CUDNN 9.0 (it would be better to use CUDNN 8.7)

Then I do 3 copy steps

Copy step 1:

Copy all files from C:\Program Files\NVIDIA\CUDNN\v9.0\bin\11.8\ to C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.8\bin\

Copy step 2:

Copy all files from C:\Program Files\NVIDIA\CUDNN\v9.0\include\11.8\ to C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.8\include\

Copy step 3:

Copy all file from C:\Program Files\NVIDIA\CUDNN\v9.0\lib\11.8\x64\ to C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.8\lib\x64\

Now I add the environment variables like this

Restart my computer and open matlab again. I then type

gpuEnvObj = coder.gpuEnvConfig;
gpuEnvObj.BasicCodegen = 1;
gpuEnvObj.BasicCodeexec = 1;
gpuEnvObj.DeepLibTarget = 'cudnn'; % it can be changed to 'tensort'
gpuEnvObj.DeepCodeexec = 1;
gpuEnvObj.DeepCodegen = 1;
results = coder.checkGpuInstall(gpuEnvObj)

Another test is to use gpucoderSetup on the matlab command prompt

You can then run checks

Answer 4

[Jaya Shankar]

Hi Susama

Looks like the requisite environments for CUDNN and TENSORRT are not set correctly on your windows.

These environment variables should point to the location on your computer where these libraries were installed as described here

https://www.mathworks.com/help/gpucoder/gs/setting-up-the-toolchain.html

You can confirm if they are set correctly by running the following commands in MATLAB session

>> getenv('NVIDIA_CUDNN')

>> getenv('NVIDIA_TENSORRT')

If the above commands return empty , make sure to set the variables through you Windows's environment variable settings found via Control Panel ->System and Security->System->Advanced System settings.

Jaya

[Susama Bagchi]

Hi Jaya,

As per your given link, I already set the path and that can be seen from the below image. But when I tried your given code

>>getenv('NVIDIA_CUDNN')

I obtained below result:

ans

0x0 empty char array

So I would like to request you to let me know what variables I need to set in control panel.

Looking forward to hearing from you.

With regards

Susama

Answer 5

[Sourabh Kondapaka]

Ensure that cuDNN library is installed in the correct directory.

Check Nvidia’s official documentation for installing in Windows :

https://docs.nvidia.com/deeplearning/sdk/cudnn-install/index.html

These environment variables should point to the location on your computer where these libraries were installed as described here:

https://in.mathworks.com/help/gpucoder/gs/setting-up-the-toolchain.html

Important Note: The Operating system ( in your case , Windows 10) only uses environment variables which were made available when the system has started. So in order for windows 10 to be able to start using the new environment variables which you had just set you need restart your system. In other words, in order to use the new or edited environment variables you will need to restart your system.

[Susama Bagchi]

Hi,

I am using Windows 10 and MATLAB 2019b. My system has NVIDIA GEFORCE RTX 2060.

From MATLAB customer care, it hase been informed that only CUDA 10.1 is compatible with MATLAB 2019b. So I installed the same along with CUDA toolkit. I found from the below NVIDIA link that cuDNN is only a zip file containing 3 folders:

https://docs.nvidia.com/deeplearning/sdk/cudnn-install/index.html#install-windows

After unzipping, I moved the main folder as shown below:

But I am still facing problem.

Susama

[Sourabh Kondapaka]

The below link contains the approach for installing cudnn and then setting the environment variables in the Control Panel.

https://docs.nvidia.com/deeplearning/sdk/cudnn-install/index.html#install-windows

After this step it would be better to restart your PC and then check if the environment variables are available to Windows OS by following Jaya Shankar's Answer above.

Steps:

1) Follow the instructions in the above link for installing cuDNN library.

2) Set the environment variables by going to the control panel. Steps for this are also available in the same link and can be easily found online.

3) Restart your PC then open Matlab and check if the OS is able to access the environment variables by following Jaya Shankar's answer.

[Ritesh Panday]

Hi Sourabh, i've followed the steps on Nvidia's webpage, but the environment variable for cuDNN doesn't seem to autoset. I even tried adding it myself, but Matlab is not detecting it. As well, i'm getting the following error:

Error using coder.checkGpuInstall (line 33)

One or more of the system checks did not pass, with the following errors ...

cuDNN Environment: (Unable to find cuDNN header files in directory 'C:\Program Files\NVIDIA GPU Computing

Toolkit\CUDA\cuda\include'. Check that the cuDNN headers are installed with the specified cuDNN SDK.)

Error in PedestrianDetectionExample (line 31)

coder.checkGpuInstall(envCfg);

I've tried all sorts of troubleshooting available in matlab forums, but they're not helping. This is really urgent, so i'd appreciate any help. Thank you!

[Susama Bagchi]

Hi Ritesh,

It is better to contact the MATLAB support services. They are helpful and can resolve your issue.

Answer 6

[Stefano Marrone]

Hi, same problem here.

coder.checkGpuInstall

Compatible GPU : PASSED

CUDA Environment : PASSED

Runtime : PASSED

cuFFT : PASSED

cuSOLVER : PASSED

cuBLAS : PASSED

cuDNN Environment : PASSED

Basic Code Generation : FAILED (Test GPU code generation failed with the error 'emlc:compilationError'. View report for further information: View report)

ans =

struct with fields:

gpu: 1

cuda: 1

cudnn: 1

tensorrt: 0

basiccodegen: 0

basiccodeexec: 0

deepcodegen: 0

deepcodeexec: 0

tensorrtdatatype: 0

profiling: 0

Did you solve it?

[Susama Bagchi]

Please contact MATLAB support services.

[Sipho Frank Sithole]

Good day as per the above conversation I am also having difficulty using the GPU coder on matlab, I did run matlab as administrator but still getting the same results. My GPU is GTX 1070, downloaded the latest cuda v11,1 and latest cudnn. However the GPU coder returns failed massage same as above comments "Basic Code Generation : FAILED (Test GPU code generation failed with the error 'emlc:compilationError'. View report for further information: View report)" Your help will be highly appreciated. I did download Visual studio 2019, I see that you have to link certain files but do not really know how. I am fairly new and working on a deep learning project.

[Ali Al-Saegh]

Hello,

Please anyone solved this problem, please help me.

Basic Code Generation : FAILED (Test GPU code generation failed with the error 'emlc:compilationError'. View report for further information: View report)

[Lo jungle]

also help me too

Answer 7

[Sehairi K.]

Hello

try this

% specify the CUDA install directory, you must have already copied cudnn files there
setenv('NVIDIA_CUDNN','C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.3')
% specify the TensorRT path
setenv('NVIDIA_TENSORRT','C:\Program Files\NVIDIA GPU Computing Toolkit\TensorRT-8.0.3.4.Windows10.x86_64.cuda-11.3.cudnn8.2\TensorRT-8.0.3.4')
coder.checkGpuInstall('full')

gpu: 1

cuda: 1

cudnn: 1

tensorrt: 1

basiccodegen: 1

basiccodeexec: 1

deepcodegen: 0

deepcodeexec: 0

tensorrtdatatype: 1

profiling: 1

Answer 8

[muhammad ahmad]

how did you resolve deepcodegen and deepcodeexec . do i need to install opencv for it

if so how can i do this

[Hariprasad Ravishankar]

Hi Muhammad,

You do not need to install OpenCV. You can resolve deepcodegen and deepcodeexec by downloading NVIDIA CuDNN and NVIDIA TensorRT libraries and setting the environment variables 'NVIDIA_CUDNN' and 'NVIDIA_TENSORRT' to point to the install folder.

Here is the documentation page of the config settings to test deepcodegen and deepcodeexec using coder.checkGpuInstall, for reference

https://www.mathworks.com/help/gpucoder/ref/coder.gpuenvconfig.html#mw_8adcfc13-a92d-4963-a267-bcb81ed268ee

Answer 9

[Kunal Khandelwal]

i am tying to use GPU Coder.

mex -setup C++

MEX configured to use 'MinGW64 Compiler (C++)' for C++ language compilation.

>> gpuEnvObj = coder.gpuEnvConfig;

gpuEnvObj.BasicCodegen = 1;

gpuEnvObj.BasicCodeexec = 1;

gpuEnvObj.DeepLibTarget = 'tensorrt';

gpuEnvObj.DeepCodeexec = 1;

gpuEnvObj.DeepCodegen = 1;

results = coder.checkGpuInstall(gpuEnvObj)

Compatible GPU : PASSED

CUDA Environment : PASSED

Runtime : PASSED

cuFFT : PASSED

cuSOLVER : PASSED

cuBLAS : PASSED

cuDNN Environment : PASSED

TensorRT Environment : PASSED

Host Compiler : FAILED (Unable to find a supported C++ host compiler. For a list of supported compilers, see supported compilers)

results =

struct with fields:

gpu: 1

cuda: 1

cudnn: 1

tensorrt: 1

hostcompiler: 0

basiccodegen: 0

basiccodeexec: 0

deepcodegen: 0

tensorrtdatatype: 0

deepcodeexec: 0

I have alreadyinstalled MinGW and Visual Studio

Can I get some help