using Legacy GPU for display and new GPU for CUDA computation

I have been working on setting up my system(Ubuntu 16.04) to use the included Quadro FX 3800 for dsiplay via display port and a GTX 750 Ti for CUDA, specifically tensor flow. initially I updated teh driver to nvidia-387 and that enabled my 750 ti to work in TF while obviously loosing the FX 3800. When i install the legacy nvidia-340 driver my system is able to recognize both cards and provide stats in the Nvidia xserver settings, as well as display over both cards although the 750 Ti does not function properly in display. This is not a problem because I dont intend to use this card for display.
The problem occurs when i try to run a tensorflow cuda program, the following error occurs:

" cannot open shared object file: No such file or directory

My assumption is this was included in the nvidia-387 driver which was purged to create the 340.

My question is will a fresh install of CUDA to the 340 version fix this, OR is there an alternative way to use both drivers, OR is this setup impossible and I should just reinstall the 387 and ru ndisplay and calc off the 750 Ti with a hit to my memory. Thank you in advance!

no, it simply won’t work.
You cannot have 2 drivers loaded.
The 387 driver is needed for CUDA 9 and will not work with the old GPU.
The 340 driver will not work with the newer GPU or CUDA 9.