Flight Log
GPU
Keeping a DGX Spark alive when unified memory runs out
Why an over-committed model load livelocks a GB10 instead of failing, why MemAvailable and the OOM killer miss it, and the guard and settings I use now.
Sharing the DGX Spark GPU with MicroK8s
Setting up MicroK8s, the NVIDIA GPU Operator, and GPU time-slicing on my DGX Spark, including the containerd paths and memory-sharing tradeoffs.