ChecklistBilling & plansPerformance & optimizationAdministration
Cost optimization checklist
Keep compute spend predictable without giving up throughput: what to check weekly, how to stop paying for idle clusters, and where to right-size.
1Know what you are spending
Check usage against plan limits weekly
The dashboard shows current usage against your plan. A weekly glance turns a month-end surprise into a Tuesday adjustment.
Attribute spend to workloads
Know which jobs dominate your compute. Optimizing the wrong one is effort with no return on the bill.
2Stop paying for idle
Shut down clusters when the run finishes
A cluster left running after a job completes costs the same as one doing work. This is the single largest avoidable line item.
Do not leave interactive sessions open overnight
A session holds resources whether or not you are typing. Close it when you stop for the day.
3Spend efficiently when you do spend
Size the cluster to the work, not to the maximum
Past the point where per-task overhead approaches per-task work, extra workers add cost without reducing wall-clock time.
Request GPUs only where they help
GPU nodes cost more. Reserve them for the tasks that genuinely benefit, and leave the rest on CPU.
Fix I/O before scaling out
Scaling an I/O-bound job multiplies waiting. Reducing read volume is almost always cheaper than adding workers.