GPU support in containers in MLOps - Practice Problems & Coding Challenges

Challenge - 5 Problems

🧠 Conceptual

intermediate

Understanding GPU Access in Docker Containers

Which of the following Docker run commands correctly enables GPU support for a container?

A docker run --enable-gpu nvidia/cuda:latest

B docker run --device=/dev/gpu nvidia/cuda:latest

C docker run --gpus all nvidia/cuda:latest

D docker run --gpu-enabled nvidia/cuda:latest

💡 Hint

Think about the official Docker flag that grants GPU access.

The correct flag to enable GPU support in Docker is --gpus. Option C uses --gpus all which grants access to all GPUs. Other options use incorrect or non-existent flags.

💻 Command Output

intermediate

Output of GPU Detection Inside Container

What is the output of running docker run --gpus all nvidia/cuda:11.0-runtime nvidia-smi on a system with one NVIDIA GPU?

A Displays GPU details including model, driver version, and usage stats

B Command not found error

C No devices were found

D Permission denied error

💡 Hint

Consider what nvidia-smi shows when GPU is accessible.

When the container has GPU access, nvidia-smi shows detailed GPU info. If GPUs are not accessible, it shows errors or no devices.

❓ Configuration

advanced

Configuring NVIDIA Container Toolkit

Which configuration file must be modified to enable the NVIDIA Container Toolkit to automatically provide GPU support for Docker containers?

A /etc/docker/daemon.json

B /usr/local/nvidia/toolkit.conf

C /etc/nvidia-container-runtime/config.toml

D /etc/docker/docker-compose.yml

💡 Hint

Think about Docker daemon configuration files.

The NVIDIA Container Toolkit requires adding a runtime entry in /etc/docker/daemon.json to enable GPU support automatically. Other files are unrelated or do not exist.

❓ Troubleshoot

advanced

Diagnosing GPU Access Failure in Container

You run docker run --gpus all nvidia/cuda:11.0-base nvidia-smi but get the error nvidia-smi: command not found. What is the most likely cause?

A The <code>--gpus</code> flag is misspelled

B The host GPU drivers are not installed

C The Docker daemon is not running

D The container image does not include the <code>nvidia-smi</code> tool

💡 Hint

Consider what is inside the container image.

If nvidia-smi is missing, it means the container image lacks the tool. The nvidia/cuda:11.0-base image is minimal and may not include it. GPU drivers on host or Docker daemon status would cause different errors.

🔀 Workflow

expert

Steps to Enable GPU Support in Kubernetes Pods

What is the correct order of steps to enable GPU support for a Kubernetes pod?

A 3,1,2,4

B 1,2,3,4

C 2,1,3,4

D 1,3,2,4

💡 Hint

Think about preparing nodes first, then enabling Kubernetes to manage GPUs, then using them in pods.

First, install drivers on nodes. Then deploy the NVIDIA device plugin so Kubernetes can schedule GPUs. Next, create pods requesting GPUs. Finally, verify GPU access inside pods.