Failed to initialize GPU BLAS library

At the start of training my CNN I get the following error:

options = trainingOptions('rmsprop',...

'MaxEpochs', 10,...

'ValidationFrequency', 50,...

'Verbose',true, ...

'InitialLearnRate', 1e-3 ,...

'MiniBatchSize', 64,...

'ExecutionEnvironment', 'gpu',...

'ValidationData', imdsValid_rand, ...

'Shuffle', 'every-epoch', ...

'Plots', 'training-progress',...

'L2Regularization', 0.0010);

%% Train the CNN model

Net_embryo = trainNetwork(imdsTrain_rand, lgraph, options);

Error using trainNetwork (line

184)

Failed to initialize GPU BLAS

library.

Error in TrainCNNClassifier (line

157)

Net_embryo =

trainNetwork(imdsTrain_rand,

lgraph, options);

Caused by:

Error using

gpuArray/hTimesTranspose

Failed to initialize GPU BLAS

library.

Failed to initialize GPU

BLAS library.

My Execution environment is set to 'gpu'. My training progresses smoothly with this set to 'cpu' but it takes a lot of time to train. I have tried updating my NVIDIA driver (latest version)

Output of gpuDevice:

Name: 'NVIDIA GeForce GTX 1660 Ti'

Index: 1

ComputeCapability: '7.5'

SupportsDouble: 1

DriverVersion: 11.4000

ToolkitVersion: 11

MaxThreadsPerBlock: 1024

MaxShmemPerBlock: 49152

MaxThreadBlockSize: [1×3 double]

MaxGridSize: [1×3 double]

SIMDWidth: 32

TotalMemory: 6.4425e+09

AvailableMemory: 4.8578e+09

MultiprocessorCount: 24

ClockRateKHz: 1590000

ComputeMode: 'Default'

GPUOverlapsTransfers: 1

KernelExecutionTimeout: 1

CanMapHostMemory: 1

DeviceSupported: 1

DeviceAvailable: 1

DeviceSelected: 1

nvidia-smi

+-----------------------------------------------------------------------------+

| NVIDIA-SMI 471.68 Driver Version: 471.68 CUDA Version: 11.4 |

|-------------------------------+----------------------+----------------------+

| 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 ... WDDM | 00000000:01:00.0 On | N/A |

| N/A 64C P0 25W / N/A | 1217MiB / 6144MiB | 0% Default |

| | | N/A |

+-------------------------------+----------------------+----------------------+

I had excatly the same error on my Ubuntu-machine which was fixed after I did sudo apt upgrade. Howerver, I can't seem to find a solution for my Windows system. Problem is mostly a missing library or something. But I can't figure it out. Thank you for help in advance. Best

Dhruv

[Joss Knight]

There certainly have been some bugs in the 470 drivers - you might want to try downgrading your driver to a 466 or earlier. I haven't seen this particular problem outside faulty drivers or use of forward compatibility mode. You should also be sure to take the latest Update to MATLAB release R2021a, since there have been some fixes to NVIDIA libraries.

[dhruv khatri]

Thank you Joss, I will try downgrading my drivers.

[Liu Xi]

hello,I meet this problem,too.This may be due to the explosion of gpu memory.Setting parameter 'Batachsize' lower may be able to solve.