I have a PC with a GTX-580 and a GTX-590. Now I have used one card (GTX-590) for my screens, and the second card for rendering (GTX-580).
This now works fine, if I just select the GTX-580 as CUDA device.
In the CUDA selection bar, I have only the option to select either all CUDA devices, or a single CUDA device. Would it be technically possible to make a selection of all the possible CUDA devices for rendering. Basically replace the drop down list with a check box list.
I guess this could able be very useful if you have a CUDA only card (where no screen can be attached), as the rendering is slowing down the window manager a lot during rendering of all CUDA devices are selected. Also the memory configuration might be different, and from what I remember to know is that CUDA is looking at the maximum memory configuration of all cards… so if your GPU only has 1G-RAM, and your CUDA card would have 4G, only 1G would be used…
So my request/question:
is it possible to have a list of CUDA devices to select from?
Hi,
I already read a similar question elsewhere and I think the only way to do so is to disable CUDA for the cards you don’t want to select for CUDA rendering.
Anyway, I don’t get you interest in this question. You have two card, meaning you can choose either one or the other, or both for rendering.
This question concerns people that use 3 or more CUDA cards and want to use one for display only and two for render only (they only have choice between 1 card or all cards at the time). From what you say, it is not your case.
Thanks for the feedback and that this might get addressed in future.
Crus-T, well my question is more like that I can get my hands on two NVIDA-Grid-K2 cards, and in this setup I would have 5 (or even 6 if I keep the GTX-590 as display card) CUDA devices. This setup would definitely ruin the power of the GRID cards, as they have more memory as the GTX-590!
@rgischig, well, then maybe disabling CUDA on your 590 is a good option.
By the way, I never saw benchmark from the K2 in cycles. If you test the card, I’d be curious to know how well it performs.
I found a solution to the “problem”. I’ve created a new nVidia application profile (in the “Manage 3d setting” section), and selected only the cards (in my case the GRID K2) for cuda. Then in Blender you see only these cards as a GPU option. This works nice.
About the GRID K2 performance: using hte BlenchMarkScene (with the BMW in default configuration, the render time is 52.83s, which is not worldclass, but similar to the GTX-590.