Twelve follow-up labs on the same gaming-laptop cluster from Part 1 : monitoring with DCGM, sharing GPUs via time-slicing, serving models with Triton and TensorRT, scheduling under contention, and CUDA profiling with Nsight Systems.

Part 2 of the NCA-AIIO Home Lab Series#

If you already built the cluster in Part 1 (WSL2 → Docker → Kind → GPU Operator → Ollama), this article is the next step. It does not repeat the setup — it assumes that stack is running and focuses on GPU operations you will meet in production and on the NCA-AIIO exam: telemetry, validation, runtime hooks, NGC containers, inference serving, and scheduler behavior on a single GPU.

The labs include break-and-fix exercises on the GPU Operator validation chain, side-by-side runc vs nvidia runtime demos, and practical limits of WSL2 (embedded DCGM, profiling depth, no MIG on consumer GPUs).

Read the full article →