Error with cudnnGetConvolutionForwardWorkspaceSize

I get an error code CUDNN_STATUS_NOT_SUPPORTED (The combination of the tensor descriptors, filter descriptor and convolution descriptor is not supported for the specified algorithm.) when calling this API.

As far as I checked , I seemed to passing in all the parameters/descriptors correctly as per what is supported

MW_mangled_ipDesc, -> points to an input of size (NxCxHxW = 1x4x224x224)
MW_mangled_filterDesc, -> points to filter of size (OxCxHxW = 64x4x7x7)
MW_mangled_convDesc, -> has a stride of 2, padding of 3
MW_mangled_outputDesc, -> points to output of size 1x64x112x112

The problem is resolved if I decreased the kernel size to 64x4x5x5
All the formats are CUDNN_TENSOR_NCHW_VECT_C and CUDNN_DATA_INT8x4.

Could this error be thrown if I am simply running out of memory ? Has anyone run into this before ?