wintpe wrote:so unless you are on some old 2000 era hardware you should be OK with nvidia.
regards peter
That's an interesting perspective, Peter, but have you ever tried to install Nvidia CUDA support without trashing an already existing screen driver setup from another manufacturer? Don't forget that the starting point from which I am trying to find a solution is a 2016 motherboard with a 2016 APU from AMD. Blender currently requires OpenCL of at least v2.0 - the A10 does that.
What I was hoping to do was gain some small performance boost while using Blender without going through the pain of getting the GeForce GTX 960 to supply CUDA support. I was also interested to test the claim that Blender's current level of OpenCL support puts OpenCL on a par with CUDA for the first time. Admittedly, the GTX 960 would still leave the A10 standing in a render race, but that is not the point. I am not looking for an absolute performance level target, just something better than plain ol' CPU rendering.
So in the end I bit the bullet and set about trying to replicate the success I had a couple of years ago getting the CUDA system running without the video drivers. Now I have the Nvidia card in one machine, using the AMD APU to drive its three screens, and the Creative Audigy card in the other. That is a very expensive way to work, but as the Nvidia card effectively renders the short PCIE socket useless, it was the best I could come up with using existing hardware.
On the plus side, though, I can use the box with the Nvidia card for Blender via x2go using vglrun to get it to use its amdgpu-provided screen acceleration and Nvidia CUDA support. This enables me to keep the three monitors on the sound machine and not worry about kvm switches, whilst still having a happy user experience with Blender. The only downside is that I dread kernel updates which will try to update or rebuild the Nvidia kernel module and probably do untold harm to the CUDA installation.