I just ran into the same thing. My guess is this is a bug in the CoreML compiler or on-device scheduler, where it tries to put part of the network onto the Neural Engine, even though it contains convolutions that need more memory than the Neural Engine can handle.

This was recently acknowledged as an Xcode 13 bug here: https://github.com/apple/coremltools/issues/1301

Perfect, thank you for the quick reply and excellent workaround! Much appreciated!

It is unfortunate that SceneKit appears to be abandoned. It would have been nice to get for example raytracing capabilities added into SceneKit. Feels like that would have been a fairly natural fit. I have always really liked SceneKit and want to thank the Apple engineers that worked on it. It is such a well designed system, especially for getting started quickly. When using plain OpenGL back in the days, the most painful painful and frustrating part of a project was often to get to the point when you could see a first triangle (or other 3D object) on screen. SceneKit made that so much faster and better by having smart defaults for automatic camera and light configuration. I think it probably fell short in its ambition to become a game engine able to compete with Unity, Unreal etc. I think the share of full-fledged games created in SceneKit compared to those engines has got to be tiny. But for just displaying some objects in 3D, making them look great and animating them in simple ways SceneKit still is fantastic. PS: Using SCNMaterialProperty.contents from type-safety obsessed Swift always amuses and delights me 😄 That's got to be one of the most promiscuous APIs Apple offers.