promptdojo_

Cloud, GPUs, Kubernetes, and cost — step 6 of 7

Price spot with eyes open: a 48-hour training job on on-demand ($2.00/hr) vs interruptible spot ($0.60/hr) where interruptions and checkpoint restarts stretch runtime by 20%. Print the on-demand cost, the spot cost (f"spot (with {overhead:.0%} runtime): ..."), and the savings line ending "- IF the job checkpoints (ch.43)". Spot is discounted BECAUSE it can be yanked — checkpoints are what make that survivable.

full-screen editor opens — close anytime to keep reading.