To use GPUs with legacy engines, you must create a custom CUDA-capable engine image.
What kind of feedback do you have?
