Resnet50 on single node with 8 GPUs, all the parameters are default. why the result is different ?

[sdwhzh]

Hello, I use the command "python main.py -a resnet50 --dist-url 'tcp://127.0.0.1:60000/' --dist-backend 'nccl' --multiprocessing-distributed --world-size 1 --rank 0 /my_data_dir/" train and test resnet50 on a single node with 8 GPUs. But I got Acc@1 75.694 Acc@5 92.704, this is different from the result presented on https://github.com/facebookarchive/fb.resnet.torch/blob/master/pretrained/README.md (ResNet-50 error rate TOP1:24.01 TOP5:7.02). All the parameters are default. why the result is different ?