Vulkan SDK detects only 1 GPU at a time

Question

I have an Optimus notebook (meaning it has two graphics cards; one integrated and one discrete (dedicated)). Both of them support Vulkan and I can use each of them individually on my code. But there is a problem, I can't list or use them both on the same program. I must select on which GPU it runs before running my code (on Windows: right click on "myProgram.exe" -> Run with graphics processor, on Linux: ./"myProgram" for integrated GPU and primusrun ./"myProgram" for discrete) and it detects only the one I run my code on even when both GPUs are active at the same time.

My GPUs are Intel UHD Graphics 630 and NVIDIA GeForce GTX 1050 Ti. I am using LunarG Vulkan SDK 1.1.92, Intel driver version 25.20.100.6326, NVIDIA driver version 416.32 on Windows and 415.25 on Linux. As C compiler and operating system, MSVC 1912 on Windows 10 and gcc 8.2.1 on Arch Linux. Vulkan runtime and ICD loaders are installed and I get the same result on both OSs.

Here is my code:

uint32_t deviceCount;
vkEnumeratePhysicalDevices(instance, &deviceCount, NULL);
printf("Found: %d\n", deviceCount);

VkPhysicalDevice *devices = malloc(deviceCount * sizeof(VkPhysicalDevice));
vkEnumeratePhysicalDevices(instance, &deviceCount, devices);

for(uint32_t deviceIndex = 0; deviceIndex < deviceCount; deviceIndex++)
{
    VkPhysicalDeviceProperties deviceProperties;
    vkGetPhysicalDeviceProperties(devices[deviceIndex], &deviceProperties);
    printf("%s\n", deviceProperties.deviceName);
}

Here is the outputs I get:

If I run on integrated GPU:

Found: 1
Intel(R) UHD Graphics 630

If I run on dedicated GPU:

Found: 1
GeForce GTX 1050 Ti

What I am expecting is:

Found: 2
Intel(R) UHD Graphics 630
GeForce GTX 1050 Ti

And I should be able to pick either of them to use on my program without hassling with some driver level code or setting it on my OS before running. Vulkan is relatively low level API after all. Is there any way to force list all of the active GPUs on my system?

Answer 1

I'm not sure what you mean by

If I run on integrated GPU

versus

If I run on dedicated GPU"...

If you're going into the nVidia control panel and specifying a target GPU for your application, then you're seeing pretty much what I would expect... only that GPU exposed to your app.

I believe what you should be doing is setting it to auto-detect, and then ensuring that you're exporting the "Optimus Aware" flag in your binary. You should have something like this in your test code

extern "C" {
_declspec(dllexport) uint32_t NvOptimusEnablement = 0x00000001;
}

That should either cause Vulkan to see both GPUs, or force your application to automatically see only the discrete GPU, but without you having to do anything in the control panel, not sure which.