cuda.bindings.nvml.device_get_gpu_instances#

cuda.bindings.nvml.device_get_gpu_instances(intptr_t device, unsigned int profile_id)#

Get GPU instances for given profile ID.

Unlike most array-returning MIG APIs, nvmlDeviceGetGpuInstances does not accept a NULL buffer to discover the count; use the number of possible placements for the profile as a safe upper bound.

Parameters:

device (intptr_t) – The identifier of the target device.
profile_id (unsigned int) – The GPU instance profile ID. See nvmlDeviceGetGpuInstanceProfileInfo.

Returns:

Returns pre-existing GPU instances for the given: profile.

Return type:

list[intptr_t]