Design a Dockerized GPU test pipeline
Overview
This question evaluates a candidate's competency in containerized GPU-aware CI system design, covering containerization, NVIDIA/AMD driver and runtime integration, headless graphics rendering, reproducibility, security, debugging, flaky-test mitigation, and CI performance optimization.
NVIDIA
Aug 9, 2025, 12:00 AM
Software Engineer
Take-home Project
System Design
0
0
Design a Docker-Based Environment for Automated Graphics Tests on NVIDIA/AMD GPUs
Context
You need to design a reproducible, secure, and debuggable CI environment that runs automated graphics tests (e.g., Vulkan/OpenGL/EGL) in Docker on Linux hosts equipped with NVIDIA and/or AMD GPUs. The system should work headlessly and scale across CI agents.
Requirements
Describe a concrete approach covering:
- Base images to use for NVIDIA and AMD, including dev vs. runtime variants.
- Driver and runtime integration (e.g., NVIDIA Container Toolkit, ROCm/DRM), device exposure, and ICD/loader handling.
- Headless rendering strategy (EGL/Vulkan vs. Xvfb) and test harness basics.
- Image layering and caching strategy to speed builds.
- Reproducibility: version pinning, driver/toolchain alignment, and environment capture.
- Security: least-privilege containers, capabilities, device nodes, secrets management.
- Debugging inside containers: tools, logging, profiling, core dumps, validation layers.
- Handling flaky graphics tests: stabilization techniques and retry/quarantine policies.
- Measuring and reducing CI runtime: metrics to track and optimizations to apply.
Deliverables
- High-level architecture (host vs. container responsibilities; per-vendor specifics).
- Example Dockerfiles (builder vs. runner), run flags, and minimal CI runner configuration.
- A checklist of metrics and concrete actions to reduce runtime while keeping determinism.
Solution
Login required
Show