Spot pricing for GPU Servers
We are introducing spot pricing for GPU Servers, giving you access to the same high-performance hardware at a reduced cost for workloads that can tolerate interruptions.
Spot pricing is now available for NVIDIA L40S and NVIDIA B200 plans, both offered at a discount compared to standard on-demand pricing. Spot instances are identical to their on-demand counterparts in terms of hardware specifications and performance — the difference is that they can be reclaimed when capacity is needed, with a 30-second graceful shutdown notice. See Pricing for details.
Spot GPU Servers are a good fit for:
- Batch inference and data processing pipelines.
- Model training jobs that support checkpointing.
- Development, experimentation, and testing.
- Any GPU workload that can be interrupted and resumed.
Storage and IP addresses are preserved if a spot instance is stopped, so you can pick up where you left off when capacity becomes available again.
See the Spot instances page for full details on the termination mechanism, best practices, and how to get started.
