Different output of the lighting model on different graphics card

[07hyx06]

Hi, thanks for your code!

I run the following code on rtx3090 and a800 under the same conda environment, but find the results are different:

import splitsum.light as light
import torch
from torchvision.utils import save_image


envlight_path = "data/irrmaps/aerodynamics_workshop_2k.hdr"
envlight = light.load_env(envlight_path, scale=1.)

save_image(envlight.diffuse.permute(0, 3, 1, 2), "diff_a800.jpg")

Here is the result on rtx3090 (Driver Version: 515.76, CUDA Version: 11.7):

Here is the result on a800 (Driver Version: 550.54.14, CUDA Version: 12.4):

Do you hava idea on this problem?

Thanks in advance.