Operational Problems Caused by GPU Use

[ShiningFeng]

I have just come into contact with NeRF for satellite image reconstruction, so I want to try to reproduce your work, but when I follow the relevant commands of Testing and Training given in README, I have the following errors, and I want your answer:

1. Testing
Running the command of Testing, the program burst with the error: RuntimeError: CUDA error: out of memory

2. Training
When running the command of Training, the initial error of the program is missing gpu_id, but I don't see this item in your instruction, so I commented it out and entered it in main.py instead.
After modification, the program loads until Validation sanity check: 0%, with an error: RuntimeError: cuda runtime error (2): out of memory at/pytorch/aven/src/THC/THCCachingHostAllocator. cpp: 278.
Later, I tried to use multiple graphics cards, but when I saw the int type specified in parser, I commented it out and made three 3090 graphics cards instead, because I saw that you used two graphics cards in your source file.

I wonder if it is because my hardware resources are difficult to meet the requirements of the project. I would like to ask you if you have any way to solve this question
Thank you.