CUDA Memory Optimization

[bkrew]

I'm not sure if this is the correct Place to ask or request.
To my Problem, achieving a good load on my 4090 is allways a bit wonky.
CUDA has some functions to help with memory handling.
I use the functions Cuda Malloc, Cuda Stream and Memory Pinning in stable diffusion Forge.
Do you think that would be usable in Onetrainer, or does Onetrainer allready use that ?