On Mon May 20 14:31:31 2024 +0000, Zhiyi Zhang wrote:
I don't really see the issue with selecting one or the other GPU if
they are both the same model. The problem is that we need to know exactly which output is connected to which GPU for things like IDXGIFactory::EnumAdapters().
I still don't understand how this is an issue. Enumerating / changing display settings is done through the driver interfaces anyway, not through Vulkan?
I understand that there might be a difference if an application looks for the GPU capabilities of the GPU that is associated with some display output, but in that case using an equivalent GPU (with the same PCI ID) should also end up with pretty much the same result, doesn't it?
In many cases there's also only a single GPU exposed from the driver, we don't even have PCI ID, and all outputs can be seen as connected it regardless. So we need to pick one as the main GPU, and any could do the job just as well as a fake one which would simply not work at all from the vulkan perspective, no?