Error using gpuArray , Maximum variable size....

Hi Matlab users and team,

I am using CNN to train a model for image segmentation. the size of input is 4D

[84 204 154 8] and this is the batch size( I can not decrease it). the network has 70 layers with Unet structure. my computer has (64 RAM, intel i7 9800x CPU, and Titan RTX GPU), and I am using Windows 10 Enterprise with matlab 2019b. first when I run the training program, I recieved error out of memory, so I just change the type of the data from double into single. however, currently I recieve the following error

Maximum variable size allowed on the device is exceeded.

Caused by:

Error using gpuArray

Maximum variable size allowed on the device is exceeded.

So i am not sure what could be the solution: add another GPU card, increase the RAM size into 128, or any possible solution!!!!

I would be thankful for any help and advice!!!

note: I found in matlab suggest to use a tall function but I got error because I am using combine function to combine image datastore with pixel datastor( the error says combine is not supported by parallel toolbox )

[Walter Roberson]

https://www.mathworks.com/matlabcentral/answers/385273-why-do-i-get-maximum-variable-size-allowed-on-the-device-is-exceeded-error-when-running-the-sema

It appears that any one array is limited to 4 Gb.

[Joss Knight]

Somewhere you are attempting to create an array with more than 2147483647 (intmax('int32')) elements. But your input is only 21111552 elements. If you are using a layer graph, please pass it to analyzeNetwork and find the culprit layer that is outputting far too large an output.