Training ResNet fails on Intel HD 620: LLVM ERROR: SPIRV internal error: Invalid magic number

[lostmsu]

The error appears as soon as the training starts.

I am running this sample: https://github.com/losttech/Gradient-Samples/tree/master/ResNetBlock
I used TensorFlow-DirectML by replacing this line with GradientEngine.UseCondaEnvironment("tf-dx-1.x");

The same code also with TensorFlow-DirectML appears to be working fine on another machine with NVidia GPU.

[jstoecker]

Thanks lostmsu. Sounds like a driver issue, so we'll need to do some investigation.