OVH provides servers in their French data center that contain two GTX970 GPUs - although these are fairly inexpensive cards, they are fairly fast for training deep learning models. This is the only host we are aware of that provides dedicated servers that are suitable for deep learning work.
To set up a server, go to the OVH GPU servers page, and choose the 'GTX 970' option. Follow the instructions to request the server. You may be contacted by OVH and asked to provide documentation to prove that you aren't a fraudster - as a result it can take a few days for them to create your first server. So we'd suggest getting started a few days ahead of when you think you'll need your server.
Once you've logged in to your server after it's been created by OVH, run the following lines - this is necessary to switch the Linux kernel from OVH's custom kernel (that doesn't support modules, including the GPU drivers) to a generic kernel that supports everything we need:
sudo -s perl -pi -e 's/GRUB_DEFAULT=0/GRUB_DEFAULT=2/' /etc/default/grub update-grub shutdown -r now
Once it's finished rebooting, follow the instructions in our generic Ubuntu installation section.