NCP-AIO Administration Practice Question

Answer analysis

Option-by-option breakdown

For each option: why learners choose it and why it is or isn't the right answer here.

• ✗

  Kubernetes namespaces with resource quotas

  Why it's wrong here

  Kubernetes namespaces manage logical access and soft limits but do not provide hardware-level compute or memory isolation. Processes within the same namespace could still contend for GPU resources, leading to unpredictable performance latency across concurrent training workloads, making this insufficient for strict hardware-level hardware partitioning requirements.

• ✗

  NVIDIA Driver process scheduling priority

  Why it's wrong here

  Process scheduling priority influences how the operating system handles tasks but does not address physical memory or hardware unit isolation. Without MIG, multiple processes on a single GPU share the global memory pool and streaming multiprocessors, causing resource contention that cannot be resolved solely through software scheduling.

• ✓

  Multi-Instance GPU (MIG) partitioning

  Why this is correct

  MIG hardware partitions ensure that each workload receives a dedicated set of compute units and memory buffers. By physically isolating the GPU resources, you guarantee deterministic performance for each tenant, which is necessary when running sensitive or high-throughput AI training models concurrently on a single hardware accelerator.

• ✗

  Docker container CPU pinning

  Why it's wrong here

  CPU pinning only affects the interaction between the CPU and the system memory management unit. It has no impact on the GPU hardware architecture or the way the GPU handles concurrent kernel executions. Consequently, GPU-bound workloads will still experience contention for shared GPU resources like streaming multiprocessors.