It will be good, especially because current GPUs have very little preemtive capacity, with single GPU that used for GUI redraw and Cycles calculation you get laggy GUI. Get one extra NVIDIA card, select it as Computing device in User Property, and you get fast smooth GUI (on AMD by default as main device) and NVIDIA that work only as Cycles acceleration. One note, try to guess what is your your typical scene RAM requirement, Cycles cannot render scene that internal data not fit in GPU RAM. As common sense, 3GB onboard RAM is some minimal requirement, but sure there are scenes that cannot fin in that too.
Best case, get card with opportunity to return, and try to render scenes with high production subdivision quality.
There is one exception, on windows there is OpenGL ICD feature that allow to coexist different vendor OPenGL drivers, In Linux side it different, you probably need some unusual modifications to environment variables to force one card as main OpenGL device, or setup and start another Xserver just for compute device.