When this caffe integrating with the cuDNN v8.0.4, there are lots of errors breaking out in t cuDNNConvolution layer test. The details explain as follow:
make the order 'make runtest'.
Any layers else but cuDNNConvolution layer test didn't break out error.
The cuDNNConvolution layer errors break out in TestSetupCuDNN,TestCconvolutionGroupCuDNN, TestGradientCuDNN and TestSimpleConvolutionCuDNN. one of the error is that the different between compute_gradient and estimated_gradient.
When this caffe integrating with the cuDNN v8.0.4, there are lots of errors breaking out in t cuDNNConvolution layer test. The details explain as follow:
The Nvidia environment: CUDA: 11.1, cuDNN 8.0.4