I’m requesting version 1.3 and the device group creation extension. I have a GTX 960 and GTX 1060 in the same machine, with the 1060 being connected to the monitor. The specification indicates that most things cannot shared without device groups. In my case, I want to present on one GPU and render on another. The issue is that both GPUs are reported as two different device groups.
I’m on Linux with NVIDIA drivers, and have no SLI bridge, and the 1060 doesn’t seem to have an SLI port.
Well, yeah. Broadly speaking, “device groups” represent GPUs that can be used in an SLI fashion. And that almost always requires that the GPUs in question are identical. Not merely from the same GPU generation, but are two copies of the same product.
If these two discrete GPUs can’t share memory (and most non-SLI devices cannot), then you’re not going to be able to do what you’re talking about without making the CPU copy data between them. Which is impractical.