ubuntu16.04+1070ti+nvidia384 cuda8.0 not working

I have installed nvidia driver 384.111 through apt install nvidia-384. Previously I installed cuda9.1 with its 387 driver, but cannot get to GUI, thus I remove 9.1, rollback to 384 and install cuda8.0 via runfile. Then I manage to compile the samples, but deviceQuery return 35, unable to detect my 1070Ti. What’s the problem?

installing CUDA 8 via runfile is not recommended after installing a driver via apt

Follow the instructions in the linux install guide to clean out old installations. Then start over, get your installers here: http://www.nvidia.com/getcuda

and read the linux install guide carefully here: http://docs.nvidia.com/cuda/cuda-installation-guide-linux/index.html

you can find older toolkits by googling “cuda toolkit archive” and the appropriate install guide will be linked from the download page

Thank you for your advice.
Today I have removed driver nvidia-384 by apt, and then uninstall cuda by runfile. Then I reboot, press Ctrl+Alt+F1 and then install cuda again. This time I accept nvidia-375, but no opengl files because I remember it will cause to loop login. And I update-initramfs -u, then reboot. But then the GUI becomes very low definition, thus it’s not correctly displayed well on my 2K screen. Although lspci | grep -i nvidia shows my GPU, but then ls /dev | grep nvidia does not show any thing. How can I fix this?
I have tried install nvidia-384 by runfile directly, but failed. The log file says pre-script failed, and then failed to build kernel module. It’s a implicit function declaration, so I guess maybe dependency error.