The variance of PSNR results among different runs is large

[machengcheng2016]

Hi,
I follow the README.md and run experiments on the Realistic Synthetic 360 dataset as python train.py --dataset_name blender --root_dir ./data/nerf_synthetic/hotdog --N_importance 64 --img_wh 400 400 --noise_std 0 --num_epochs 16 --batch_size 1024 --optimizer adam --lr 5e-4 --lr_scheduler steplr --decay_step 2 4 8 --decay_gamma 0.5 --exp_name nerf_synthetic/hotdog --num_gpus 8.
I run on eight Tesla V100 GPUs, and install the dependencies by pip install -r requirements.
However, after I run three times, the PSNR results are 31.00, 30.63, 34.14, respectively.
Why does that happen?