How much GPU vram do I need for normxcorr2 in a large image?

I have large tiff images from svs files

each file is about 10,000x10,000 pixels and 100~300MBs.

I changed these to gpu arrays, grayscaled them with rgb2gray, and tried normxcorr2 but

an error resulted in insufficient memory when running normxcorr2. My gpu is a single rtx2080super with 8GB of vram.

However, I had no problem doing the exact same thing when I imported the tiff files with imread() only and 32GB of total system memory.

Moreover, I could do normxcorr2 with gpu array when smaller images were used(as stated in the manual).

How much memory will I need to run normxcorr2 with gpu arrary of large images?

I think my process will be much faster with gpu then cpu.