Problem with GPU and Matlab version

Hi,

The cluster we have has 4 Tesla V100 GPUS, but the Matlab version available on it is 2014b. From the link below, I can tell that they are not compatible.

https://www.mathworks.com/help/parallel-computing/gpu-support-by-release.html

Is there a way to resolve the compatibility issue between the available GPU and Matlab version? My goal is to use the GPU to solve a linear system of equations using the PCG Matlab function.

Abueidda

Answer 1

[Walter Roberson]

That table is telling you that the Volta GPUs can be used with your system, but that the first time you run something there may be a delay while the driver recompiles the libraries.

You might not get as good of a performance as you would with a later release.

You would install the latest driver -- well, the latest driver that supports your architecture, which in your case is all current releases.

If those cards had been the original Tesla architecture then R2014a would have been the limit, but they are not Tesla architecture, they are Tesla branded Volta architecture.

Answer 2

[Christabel Ajaero]

Hello,

Palese, i am having a problem training my network on my computer.....this is the error message i get when it gets to the trainNetwork line

......."Warning: while copying objects of class

'gpuArray':

'Out of memory on device. To view morw details about available memory on the GPU, use 'gpuDevice'. If the problem persists, reset the GPU by dailing 'gpuDevice(1)' "............

and my computer has GeForece MX250 under the Pascal Micro-architechure with 6.1 computing capability according to

https://en.wikipedia.org/wiki/CUDA#GPUs_supported.

pls help ....i have the done instruction from the error, still not working. Im using MAtlab version R2020a

Christabel.

[Walter Roberson]

MX250 has a maximum of 4 gigabytes of memory. If you are using Windows then the operating system will reserve up to 1 gigabytes.

The message is telling you that the array you are trying to copy in is too large to fit. If it did fit you would still have problems because operations on the array would typically need another array the same size.

If you are training a rccn then reduce your minibatch size.

[Christabel Ajaero]

Hi,

Thank you for your answer.

i tried to reduce the minibatch size from the default 128 to 50, still the error persisted. what minibatch size do you suggest pls? i am traning a cnn but would still train rcnn later