Question 21
UnclassifiedAfter creating GPU instances using 'nvidia-smi mig -cgi', what additional step is required before workloads can be scheduled to the MIG instances?
Correct answer: A
Explanation
After `nvidia-smi mig -cgi` creates GPU Instances, they are only the parent partition. Workloads can be scheduled only after you "Create Compute Instances (CI) within each GPU Instance," because the CI is the schedulable unit exposed to the system.
Why each option is right or wrong
A. Create Compute Instances (CI) within each GPU Instance
Under NVIDIA MIG, `nvidia-smi mig -cgi` creates only the GPU Instance (GI) partition, which is not yet a schedulable endpoint. The next required step is to instantiate Compute Instances inside each GI using the MIG compute-instance workflow (for example, `nvidia-smi mig -cci`), because the CI is the unit presented for workload placement and scheduling. Without that CI layer, the GI remains an intermediate partition and cannot accept jobs.
B. Restart the nvidia-persistenced daemon
C. Enable MIG mode at the driver level using nvidia-smi -mig 1
D. Register the instances with Fabric Manager