Can someone explain to me why this is happening

[TeKett]

iv used a 2070 for about 2 years generating things on A1111, i have been able to generate images with sd1.5 that are 512x768 with 1.4x upscale. Later i got a 4070ti when it released, been able to do 512x768 with 2x upscale. But getting more into AI i wanted more vram. Now that the 5090 has dropped 30% in price where i live, i decided to get one. 32GB vram, and doing 512x768 with 2x upscale i now run out of memory. Exact same model (sd1.5 4gb) exact same settings, only difference is the versions of A1111 over the years.

torch.OutOfMemoryError: CUDA out of memory. Tried to allocate 9.00 GiB. GPU 0 has a total capacity of 
31.36 GiB of which 1.50 GiB is free. Including non-PyTorch memory, this process has 29.47 GiB memory 
in use. Of the allocated memory 28.77 GiB is allocated by PyTorch, and 84.99 MiB is reserved by PyTorch
but unallocated.

did it seriously try to allocate 40gb or more to generate a 1024x1536 image?

[ooofest]

I realize this question is a couple of months old, but perhaps you might look here if this is still an issue:
#16818