Some problems about CUDA

[Robin-WZQ]

Hello,

I am a sophomore student from Beijing Institute of Technology. When I use GPU to run main.py I met some problems:

First, I used one GPU but it showed a memory overflow. Thus, I had to decrease the batchsize from 32 to 16 then it worked and the result was 49%.

However, when I tried to use multi-gpu by using "os.environ['CUDA_VISIBLE_DEVICES']='0,3' and model = nn.DataParallel(model)". Assuming my code was correct and I got a error "RuntineError: Given groups = 1, weight of size 512 2049 1,expected input[192,3,200704] to have 2048 channels, but got 3 channels instead".

So, my question is if the VAANET can run in multi-gpu? If the answer is yes, could you give me some materials to learn why I got this error and I will be very appreciate if you tell me where I went wrong!

Thanks!