Cuda from a Windows Service?

Is it possible to run a kernels from the windows service somehow? I found a way on a forum, that it is possible to spawn the child process using the same service executable, but the answer was giving 5 years ago. Maybe there is some way now? I cannot even use the NVAPI from the service, which would be ok too.

There is a posting on this forum from user tbradley, including source code example. I don’t think anything has chnaged.

the posting from tbradley is here:

Yes, i’v seen that. Its spwans the child process as currently logged user from a session0 windows service, but it’s not actually what i wanted to achieve, accessing the cuda (or nvapi) from session 0 (from windows service directly). Is there maybe a way i can make a IRP requests to the nvidia driver directly using winapi “DeviceIoControl/CreateFile” ? Maybe i can find somewhere the IRP codes for at least querying the information from GPU NVIDIA driver, such as gpu load, configuration, memory, temperature?