I currently have a workstation with a Titan Xp since 3 years. It currently has Ubuntu 18.04, Cuda 10.2 and Cudnn 7.6. I’m only using this workstation for deep learning (mostly Pytorch).
Today, I have ordered the Titan RTX for the 24GB of GDDR6, which will join the Titan Xp in the machine (as two independent cards for parallel experiments). In terms of the motherboard, PCI-e connection, and power supply, everything is checked on my side. However, I have some questions about nvidia drivers and cuda:
Both GPUs seem to have the same driver https://www.nvidia.com/download/index.aspx?lang=en-us . After I install the second card, I shouldn’t do anything else in that regard as I already have drivers of the Titan Xp? If no, what should I do to handle these two cards?
(if the answer of 1. doesn’t imply additional commands)
I installed the nvidia drivers automatically with CUDA: apt install cuda. Is there a difference (in terms of performance for deep learning) between the ones with the package or the very last version available on nvidia.com?
Would having two GPUs of different types have an effect on the performance of the GPUs as they might not have the same compute capability?
Thank you very much for your help!