Which is better OpenCl (AMD) or CUDA (Nvidia) for video rendering and processing?

Question

GPGPU

Hello!
I assembled a new computer for myself (i7-4770, 32gb ram), but I haven’t bought a vidyukha yet, I haven’t picked it up. The computer is planned to be used for:
- Photoshop (accelerated through OpenCL, but here the CPU is likely to be enough)
- After Effects (accelerated by CUDA for raytracing rendering, I did not understand if the effects / filters are accelerated)
- Cinema 4d (built-in rendering is only CPU, third-party, which are screwed to C4D, there are both OpenCl and CUDA, I haven’t decided on a specific renderer yet)
- e-on Vue (CPU only, but suddenly someone knows something about GPU rendering for it)
On vidyuhu I have about 13t.r. (maybe a little more). I would like to find the most universal solution not only for the above applications, but also for related ones, for example, Premiere Pro (it seems that OpenCl speeds it up better), for other 3d. You never know what will change ...
Nvidia seems to be more universal, there are both CUDA and OpenCL. But, as far as I googled, OpenCl on Nvidia drains AMD a lot.
Although, somewhere I came across that this is due to an insufficiently optimized driver / compiler. Can we hope that in the future OpenCL on Nvidia will be better? I also read that for rendering on a vidyuhi it is necessary that the whole scene fit into its memory, so it is necessary that the memory be at least 4 GB. Cards from AMD under this condition are cheaper.
What do you advise?
Thank you!

Answer 1

If you plan to take it for rendering, then solutions from AMD definitely win in terms of price-performance ratio. In addition, OpenCL seems to me personally more promising, due to its open source nature.

Answer 2

First of all, pay attention to the bus width (the more bits - the better) and to the amount of memory in the second place. The wider the bus, the faster data will be transferred to the card and back.

Answer 3

I don't like AMD for their buggy drivers. Because of them, something is constantly, but does not work. They keep promising that they will fix it in the next version. And so the last 15 years. So I would choose NVIDIA. Plus NVIDIA supports both CUDA and OpenCL.
And, as far as I know, NVIDIA was leaking AMD in integer calculations (that's why they mined everything on AMD), but it seems like on the new architecture (starting from the 750th) NVIDIA corrected this assumption and the GeForce 750Ti became the most efficient card for mining (the best ratio integer performance / power consumption). Well, in floating point calculations, NVIDIA has never been inferior.