MESA_VK_DEVICE_SELECT !1766
Amm, i might be wrong, but its actually not solving the problem of systems with two identical gpu.
$ MESA_VK_DEVICE_SELECT=list vkcube
selectable devices:
GPU 0: 1002:687f "AMD RADV VEGA10 (LLVM 10.0.0)" discrete GPU
GPU 1: 1002:687f "AMD RADV VEGA10 (LLVM 10.0.0)" discrete GPU
With Alex`s solution, at least i can specify whic gpu i want to use by vulkan GPUid, since the PCI ids are gonna be identical. ANd in my case vulkan, for some weird reason label the primary as gpuid1 and the secondary as gpuid0, so those apps, where his solution isnt working (32bit games, all ive tried yet to figure out) this wont help either. What can help is just simply be able to tell the system, to use the other gpu by selecting that id manually. After the first setup, it didnt (the gpuids) changed in the last half year, even after complete reinstall.
And yes, based on this: vulkaninfo
VK_LAYER_AEJS_DeviceChooserLayer (Device chooser layer) Vulkan version 1.1.121, layer version 1:
Layer Extensions: count = 0
Devices: count = 2
GPU id = 0 (AMD RADV VEGA10 (LLVM 10.0.0))
Layer-Device Extensions: count = 0
GPU id = 1 (AMD RADV VEGA10 (LLVM 10.0.0))
Layer-Device Extensions: count = 0
VK_LAYER_MESA_device_select (Linux device selection layer) Vulkan version 1.1.73, layer version 1:
Layer Extensions: count = 0
Devices: count = 2
GPU id = 0 (AMD RADV VEGA10 (LLVM 10.0.0))
Layer-Device Extensions: count = 0
GPU id = 1 (AMD RADV VEGA10 (LLVM 10.0.0))
Layer-Device Extensions: count = 0
its not possible to see which is which, but on the vega56-64 and furys there is a feature, what AMD should keep on every high end gpu they make (GPUTach). And based on that (and sometimes the amount of introduced latency caused by copying the images back and forth) i know, GPUid 0 is my second gpu in pci bus 0e:00.0 and the primary gpu is GPUid 1 in 0b:00.0 .