Using GPU during dataloader + Multi-GPU training

[iqbalmohomed]

Hello,

I have a use case that is not that obscure (need to use GPU during data loading phase; using CPU is just not fast enough). I am also employing Multi-GPU training. Have machines that have 3,4 and 8 GPUs. I was hoping to dedicate one to the data preprocessing, and the rest for training. However, I have not been able to find any examples that support this type of setup. Is this possible currently? Would appreciate any pointers.