sorry that this question is more general and not only related to PyTorch, but I think here I find the experts who can give me an advice
I would like to build a new server for DeepLearning.
This shall be used mainly for CNNs, Mask RCNNs and GANs.
Initially I thought of buying 4x RTX 2080TI. As I know I could combine two of these with a NVLink-Bridge which would give me two times 22 GB. But I could not get 44 GB.
For nearly the same price as for 4x RTX 2080TI I could buy e.g. only one Quadro RTX 8000 with 48 GB.
I just read that 11 GB should be enough for the cases I described whereas for example for NLP a huge amount of GPU memory might be necessary.
Is this correct?
What is your experience?
How much memory would you like to have today for the described cases?
Thank you very much