Error using parallel.g​pu.CUDADev​ice.hBuild An unexpected error occurred trying to retrieve CUDA device properties. The CUDA error was: CUDA_ERROR_UNKNOWN

Hello I am getting error while running nueral network programs the error is "Error using parallel.gpu.CUDADevice.hBuild An unexpected error occurred trying to retrieve CUDA device properties. The CUDA error was: CUDA_ERROR_UNKNOWN" few days back it used to run properly from 2 days it is displaying this error. even i tried updating drivers no use. please help me .

[Joss Knight]

Please give MATLAB version, device, system information, and output of nvidia-smi.

[mohammed]

Error

Training throws the error 'CUDA_ERROR_UNKNOWN'

Description

Sometimes, the GPU throws this error when it is being used for both compute and display requests from the OS.

Possible Solution

Try reducing the mini-batch size using the 'MiniBatchSize' option of trainingOptions. If reducing the mini-batch size does not work, then in Windows®, try adjusting the Timeout Detection and Recovery (TDR) settings. For example, change the TdrDelay from 2 seconds (default) to 4 seconds (requires registry edit).

Answer 1

[pradeep pc]

select high performance option for MATLAB on NVIDIA control panel to solve. It works :)

[HM]

This did work for me. Thank you.

[Zaher Babakrnejad]

me too !!!

[Kevin dos Reis Vezo]

Thank you!

[Nikita Belousov]

Помогло

Answer 2

[Pramod Rao]

Matlab version 2017a , Lenovo Laptop Windows 10 ,4gb ram,1gb Graphic card , nvidia GEFORCE GT610 The error is

[Walter Roberson]

Please run nvidia-smi outside of MATLAB and include the results.

[SATYAM KUMAR CHOURASIYA 16MCA1038]

please check this and let me know , i'm facing same issue!

[SATYAM KUMAR CHOURASIYA 16MCA1038]

it say not supported?

[Joss Knight]

These mobile graphics chips are very flaky for CUDA execution because of their power requirements. Try rebooting; also, go to the NVIDIA Control Panel, Manage 3D Settings, then ensure that in Program Settings, when MATLAB is selected, the High-performance NVIDIA processor is the preferred graphics processor.

[mohammed]

Error

Training throws the error 'CUDA_ERROR_UNKNOWN'

Description

Sometimes, the GPU throws this error when it is being used for both compute and display requests from the OS.

Possible Solution

Try reducing the mini-batch size using the 'MiniBatchSize' option of trainingOptions. If reducing the mini-batch size does not work, then in Windows®, try adjusting the Timeout Detection and Recovery (TDR) settings. For example, change the TdrDelay from 2 seconds (default) to 4 seconds (requires registry edit).

Answer 3

[M. Mubashir Hussain]

I am getting the following error when I try to train my network : Error using parallel.gpu.CUDADevice.hBuild

An unexpected error occurred trying to retrieve CUDA device properties. The CUDA error was:

CUDA_ERROR_UNKNOWN

I an using Nvidia Geforce GTX 1050 TI GPU on my Windows 10 Laptop. When I restart Matlab, error goes away but reappears when I train network again. I have the latest Nvidia driver. I feel that I hve started getting this error after I installed Cuda toolkit 9 and CuDNN libraries to make tensorflow for GPU work?

Interestingly my GPU is dectable using the various MATLAB commands.

[mohammed]

Error

Training throws the error 'CUDA_ERROR_UNKNOWN'

Description

Sometimes, the GPU throws this error when it is being used for both compute and display requests from the OS.

Possible Solution

Try reducing the mini-batch size using the 'MiniBatchSize' option of trainingOptions. If reducing the mini-batch size does not work, then in Windows®, try adjusting the Timeout Detection and Recovery (TDR) settings. For example, change the TdrDelay from 2 seconds (default) to 4 seconds (requires registry edit).

Answer 4

[M. Mubashir Hussain]

Any replies please? The only solution is to restart MATLAB everytime this error occur? Any other solution please?

[mohammed]

Error

Training throws the error 'CUDA_ERROR_UNKNOWN'

Description

Sometimes, the GPU throws this error when it is being used for both compute and display requests from the OS.

Possible Solution

Try reducing the mini-batch size using the 'MiniBatchSize' option of trainingOptions. If reducing the mini-batch size does not work, then in Windows®, try adjusting the Timeout Detection and Recovery (TDR) settings. For example, change the TdrDelay from 2 seconds (default) to 4 seconds (requires registry edit).

Answer 5

[M. Mubashir Hussain]

I am now getting this error:

Error using gpuDevice

An unexpected error occurred trying to retrieve CUDA device properties. The CUDA error was:

CUDA_ERROR_UNKNOWN

[Joss Knight]

I think you need to upgrade MATLAB, it may be that there's a problem with using this newer NVIDIA driver with an older cuDNN. You can get R2018b right now, or wait a few weeks for R2019a.

[mohammed]

Error

Training throws the error 'CUDA_ERROR_UNKNOWN'

Description

Sometimes, the GPU throws this error when it is being used for both compute and display requests from the OS.

Possible Solution

Try reducing the mini-batch size using the 'MiniBatchSize' option of trainingOptions. If reducing the mini-batch size does not work, then in Windows®, try adjusting the Timeout Detection and Recovery (TDR) settings. For example, change the TdrDelay from 2 seconds (default) to 4 seconds (requires registry edit).

[Oscar]

I have the same error, what works for me is using the command: gpuDevice(1)

I was hoping to find a better alternative to my solution, but maybe there's none other.