Using multiple GPUs in training or testing

[xvyaward]

Hi! Currently, I'm trying to enhance multiple images sequentially with just one-time load of the SRGAN_g network. However, when I look nvidia-smi, it says SRGAN program(python) is using all of my gpus, but I think only GPU0 is fully working on it. (By Temperature, power usage, and GPU-utill %)

I have found the code with tf.device('/device:GPU:1'): on google , but I'm not sure where to put it.

So my questions are

1. Is SRGAN fully using multiple GPUs?
2. If not, how can I use multiple GPUs for inferencing? I'm working on multiple images, so giving image1,5,9 ... to GPU0 and image2,6,10 ... to GPU1 is one way to process in parallel.
3. Is there any methods to speed up testing? (without upgrading HWs..)

Thanks for reading this!