> I hope this will enable open-source hardware-accelerated image upscaling That's already a thing (especially with AMDs FSR already being open source), but reg

> I hope this will enable open-source hardware-accelerated image upscaling That's already a thing (especially with AMDs FSR already being open source), but regardless keep in mind that DLSS 2.0+ requires engine support because it also uses things like motion vectors. Retrofitting this into old games is possible, but likely non-trivial, and would require the game to be adjusted it's not something the driver can just do. > However, recent NVIDIA GPUs contain dedicated silicon to perform this type of upscaling, which I'm really hoping will make things even faster! There's no dedicated silicon for DLSS. It's "just" the tensor cores, which are programmable and is something PyTorch already has access to. CUDA support for those have been around for a while after all: https://developer.nvidia.com/blog/programming-tensor-cores-cuda-9/ https://developer.nvidia.com/blog/programming-tensor-cores-c...