You might want to compare your render needs with the list of things OpenCL rendering can’t do (compared to CUDA): Here. Add to that being unable to use HDR textures with OpenCL.
If you can live without all those features, AMD GPUs might be worth a shot. But the one million dollar question is: Will AMD GPUs still perform so well after all these features have been added to OpenCL rendering?
It is my understanding that for cycles rendering it has to be Nvidia for CUDA computing. I just built a new computer with GTX 970 and so far it’s pretty bangin’ for my buck. I did get a SLI compatible mobo so that I can add another at some point.
SLI is neither needed nor advised for GPU rendering, btw. Cycles uses all available GPUs anyway, but combining them via SLI will create so much overhead that you lose up to 50 % in performance.
And that’s completely true. But there is a difference between having two cards in your machine and running those two cards in SLI mode.
SLI is a gaming thing that more or less makes the two cards run as one. GPU computing does not need that and runs better without. Even without SLI it will still use all available cards in your machine.
Since SLI is not taken really something that benefits Cycles would having different makes and models in a system affect it then like if i had a 770 and a 970 in one system ? because i know with Crossfire there is no issue but you cant with SLI,
I remember with an older system i used to render in cycles using two different graphics cards does it bottle neck anything at all or even let you render with both if they are not the same model with Nvidia ?
Since SLI is not taken really something that benefits Cycles would having different makes and models in a system affect it then like if i had a 770 and a 970 in one system ? because i know with Crossfire there is no issue but you cant with SLI,
You can have different cards but the graphics memory will be limited to the card with the least
That’s the part that gets me, Technically the memory would be the same for both cards its all the other specs that change.
So having two cards wont help with running out of memory when rendering with GPU or will it give me a total of 8GB with two 4gb cards?
Is the render speed determined by the Memory or the clock speed or number or Cuda Cores and in that case would be be throttled by the slower one or one with the least amount of cores ?.
So having two cards wont help with running out of memory when rendering with GPU or will it give me a total of 8GB with two 4gb cards?
No, scene has to fit in every card.
Is the render speed determined by the Memory or the clock speed or number or Cuda Cores and in that case would be be throttled by the slower one or one with the least amount of cores ?.
Mostly Cuda Cores, provided using the same series because you cant compare GTX 500 with GTX 900 cores.
If one card has 2048 Cores and the second 1024 you will get ~50% more performance with both cards.