PyTorch version compatibility with GPU devices

Hello, I am having issues with compatibility between PyTorch versions / GPU devices / operating systems. Specifically, I am training and saving a neural network on a GPU device and then loading it to a different device (different GPU) with a different PyTorch version - this results in the neural network not being loaded properly. In addition, I am also training the neural network on Linux, while I load it on Windows (not sure if this can also cause problems).

Could anyone please direct me to any documentation online mentioning which GPU devices are compatible with which PyTorch versions / operating systems?

Below are the detailed information on the GPU device names and PyTorch versions I used, which I know for sure that definitely are not compatible.
Training: GPU device name: GeForce GTX 1080 Ti; PyTorch version 1.5.1; Linux.
Loading: GPU device name: GeForce RTX 2070 with Max-Q Design; PyTorch version: 1.3.0; Windows.

Thanks a lot in advance!

Could you explain what “not being loaded properly” means?
Are you checking the parameters before saving and after loading and see differences?
Also, note that forward-compatibility is not supported. I.e. saving a model in a newer PyTorch release and loading in an older might fail (the other way should work), so could you update both PyTorch installation to the latest release?