NCA Containerization & NGC Catalog 3 — Questions and Answers
Question 1: What is the function of NVIDIA's 'nvidia-container-runtime' in a containerized GPU environment?
- It replaces the Linux kernel's GPU scheduling module
- It is an OCI-compliant runtime hook that injects GPU capabilities into containers (Correct answer)
- It provides a Python API for managing container lifecycles
- It compiles CUDA code inside the container at startup
Correct answer: It is an OCI-compliant runtime hook that injects GPU capabilities into containers
nvidia-container-runtime is an OCI-compliant container runtime that hooks into the standard container lifecycle to inject GPU driver libraries and devices into containers.
Question 2: When authenticating to the NGC container registry via Docker, which registry hostname is used?
- registry.nvidia.com
- nvcr.io (Correct answer)
- hub.ngc.nvidia.com
- docker.ngc.io
Correct answer: nvcr.io
NGC container images are hosted at nvcr.io, so Docker login and pull commands must reference this hostname.
Question 3: Which environment variable is set inside NGC containers to indicate the minimum required CUDA driver version?
- CUDA_DRIVER_MIN_VERSION
- NVIDIA_REQUIRE_CUDA (Correct answer)
- NGC_CUDA_REQUIREMENT
- CUDA_VISIBLE_DEVICES
Correct answer: NVIDIA_REQUIRE_CUDA
The NVIDIA_REQUIRE_CUDA environment variable specifies the minimum CUDA and driver version required for the container to run correctly.
Question 4: In an NGC container, what does the environment variable CUDA_VISIBLE_DEVICES control?
- The list of CUDA kernels allowed to execute
- Which physical GPUs are exposed to the container's processes (Correct answer)
- The CUDA memory allocation limit per process
- The GPU clock speed ceiling inside the container
Correct answer: Which physical GPUs are exposed to the container's processes
CUDA_VISIBLE_DEVICES restricts which host GPU devices are visible and accessible to processes running inside the container.
Question 5: What distinguishes an NGC 'AI Workflow' from a standard NGC container image?
- AI Workflows are bare-metal installers, not containers
- AI Workflows bundle end-to-end reference pipelines including data preparation, training, and inference scripts (Correct answer)
- AI Workflows contain only inference engines without training code
- AI Workflows are restricted to NVIDIA-owned hardware only
Correct answer: AI Workflows bundle end-to-end reference pipelines including data preparation, training, and inference scripts
NGC AI Workflows are complete reference implementations that include all steps from data preprocessing through model training and deployment, not just a runtime environment.
Question 6: Which Kubernetes device plugin is required to schedule GPU resources in a Kubernetes cluster running NGC workloads?
- NVIDIA GPU Feature Discovery
- NVIDIA GPU Operator
- NVIDIA k8s-device-plugin (Correct answer)
- NVIDIA MIG Manager
Correct answer: NVIDIA k8s-device-plugin
The NVIDIA k8s-device-plugin exposes GPU resources to the Kubernetes scheduler, allowing pods to request GPU allocations via resource limits.
Question 7: What is the role of the NVIDIA GPU Operator in a Kubernetes environment?
- It manages GPU pricing and billing across cloud providers
- It automates the deployment and management of all NVIDIA software components needed for GPU-accelerated Kubernetes workloads (Correct answer)
- It provides a GUI for monitoring GPU utilization across nodes
- It compiles custom CUDA kernels for containerized applications
Correct answer: It automates the deployment and management of all NVIDIA software components needed for GPU-accelerated Kubernetes workloads
The NVIDIA GPU Operator uses the Kubernetes Operator framework to automate the lifecycle management of GPU drivers, container toolkit, device plugin, and monitoring components.
NCA Containerization & NGC Catalog 3
What is the function of NVIDIA's 'nvidia-container-runtime' in a containerized GPU environment?