CUDA uses dedicated hardware onboard the GPU silicon to work, these are specialised workloads and have to be specially designed for. if gaming workloads are to be advanced with higher performance then it can only be done from the ground up,. so future engines maybe. Currently all of those compute tasks are done on the CPU which is why we see so many games release with poor CPU optimisation, or engines that are only single thread biased instead of wide multithreading - The tasks they are doing are specifically CPU specialised tasks like calculating NPC pathways, number of NPCs on screen etc.
It's the generative AI stuff where the GPU takes over as well as rendering technologies like RT denoising (ray reconstruction) and DLSS etc which all use the Tensor cores and cannot work any other way.
Windows isn't even properly multithreaded or leveraging other modern API tech like DirectStorage. Imagine Windows using the full scope of CPU cores like how a properly optimised game does, how fast could Windows boot up etc. ok ~10 seconds is quick today anyway but there's no reason why Windows cannot boot instantly from cold if all these technologies were being used. There was rumour flying around Windows 12 will change a lot of under the hood stuff so maybe that will be the turning point.