Plan memory headroom and catch thermal slowdowns.
What it does
Plan training memory headroom, recover from unified-memory OOMs, and monitor thermals during long jobs on DGX Spark.
When to use it
Use before sizing a training run, after a unified-memory OOM, or when investigating thermal slowdowns during a multi-hour job.
How to use it
Give it a training run to assess; it provides a memory budget, ordered OOM remedies, or thermal diagnostics.
What you provide
- An existing project
Uses
Good to know
- cannot be undone: thermal.log
Latest review
What you need · 2
The procedures address long-running ML jobs on NVIDIA DGX Spark and its GB10 unified-memory hardware.
The thermal sampler requires nvidia-smi and exits if the command is unavailable.
Access · 8
This skill
thermal.log
Write
Overwrites the thermal log (irreversible)
nvidia-smi
Execute
Starts a background thermal sampler