Access spare compute capacity at up to 90% discount. Perfect for fault-tolerant workloads, batch processing, CI/CD pipelines, and ML training with checkpoints.
Prices vary based on demand. These are typical savings.
| Instance Type | On-Demand | Spot Price | Savings | |
|---|---|---|---|---|
| gp.small | ₹350 | ₹52 | 85% | |
| gp.medium | ₹700 | ₹105 | 85% | |
| gp.large | ₹1,400 | ₹210 | 85% | |
| gp.xlarge | ₹2,800 | ₹420 | 85% | |
| gp.2xlarge | ₹5,600 | ₹840 | 85% | |
| co.large | ₹1,200 | ₹180 | 85% | |
| co.xlarge | ₹2,400 | ₹360 | 85% | |
| mo.xlarge | ₹3,600 | ₹540 | 85% | |
| gpu.t4.1x | ₹25/hr | ₹7.5/hr | 70% | |
| gpu.a100.1x | ₹75/hr | ₹22.5/hr | 70% |
Specify your instance type and maximum price you're willing to pay.
When capacity is available and spot price is below your max, instance starts.
Use the instance just like any on-demand instance.
Get 2-minute warning via metadata service. Save state and gracefully terminate.
Access spare compute capacity at a fraction of on-demand prices.
Get notified 2 minutes before termination to save your work.
Identical hardware and performance as on-demand instances.
Handle interruptions gracefully with metadata service and webhooks.
Spot instances work best for workloads that can handle interruptions.
Process large datasets with parallelizable workloads that can checkpoint progress.
Run build and test jobs that can be retried on interruption.
Apache Spark, Hadoop, and other distributed computing frameworks.
Train models with checkpointing for fault tolerance.
Transcode video files in parallel chunks that can be reprocessed.
Stateless crawling jobs that can resume from where they left off.