In a system with multiple CUDA-capable video cards, is it possible to determine which are attached to displays?
I need this to weight (amongst other factors, like memory size) a choice of which card to use for computation.
In a system with multiple CUDA-capable video cards, is it possible to determine which are attached to displays?
I need this to weight (amongst other factors, like memory size) a choice of which card to use for computation.
Not exactly. But you can get whether the device has the watchdog timer from the deviceProp structure filled out by cudaDeviceGetProperties().
There’s no reliable way to map NVAPI device ID (or handel) to NVAPI device ID (or handle), so it won’t work.
Thanks, the watchdog should be a sufficient hint.
if you only want to know available memory, use cuMemGetInfo()