Spot Virtual Machines on Hyperstack
Spot virtual machines (VMs) offer a cost-effective way to run GPU-based workloads by utilizing Hyperstack's surplus compute capacity. This guide explains how spot VMs work, how they differ from on-demand VMs, their limitations, and the types of use cases they are best suited for.
In This Article
Overview of Spot VMs on Hyperstack
Spot VMs give you access to unused GPU capacity at a significantly reduced cost compared to on-demand VMs. They are ideal for fault-tolerant workloads that don’t require guaranteed uptime.
How to Deploy a Spot VM
To launch a Hyperstack Spot instance, deploy a VM with your desired configuration and select a flavor that includes Spot in the name—for example, A100-80G-PCIe-Spot.
For step-by-step instructions on deploying a VM in Hyperstack, see the full guide here.
Key Differences: Spot vs On-Demand
| Feature | Spot VMs | On-Demand VMs |
|---|---|---|
| Pricing | Discounted (fixed % reduction) | Standard rate |
| Availability | Not guaranteed, may be terminated anytime | Guaranteed while in use |
| Termination Notice | None | Not applicable |
| Data Persistence | No data persistence | Full persistence support |
| Public IP Charges | Standard rate | May be discounted |
Limitations of Spot VMs
Keep the following in mind when using spot VMs on Hyperstack:
- No termination notice: VMs can be deleted at any time without prior notice if capacity is reclaimed.
- No hibernation or snapshot support: Spot VMs are ephemeral.
- No boot-from-volume: All spot VMs must boot from the base image.
- No data persistence guarantees: Unsaved data will be lost if the VM is terminated.
- Standard public IP pricing applies: Discounts do not extend to public IP address charges.
- Limited flavor availability: The UI indicates which VM flavors support spot pricing.
Data stored on Spot VMs is ephemeral and will be permanently lost upon termination. It is the user’s responsibility to save important data to external storage or implement checkpointing to prevent data loss.
Please note that because spot instances can be terminated at any time, any attached volumes may become corrupted or enter an error state. Users are responsible for manually deleting these volumes if needed.
For the official policies, conditions, and disclaimers regarding Spot instances, please refer to section 4.5 of our Terms and Conditions.
Pricing
Spot VMs come with a fixed percentage discount compared to on-demand pricing, making them easy to budget for.
However, they carry a non-zero risk of termination without prior notice (no email or in-platform alerts). They are best suited for workloads that can tolerate interruptions in exchange for substantial cost savings.
When to Use Spot VMs
Spot virtual machines are best suited for workloads that are flexible, fault-tolerant, or temporary in nature. They offer significant cost savings compared to on-demand VMs, making them a strong option when performance is needed but continuous uptime is not critical.
Ideal Use Cases
- Non-critical or interruptible workloads: Tasks that can handle unexpected interruptions, such as background jobs, testing environments, or temporary development setups.
- Batch processing: Workloads that run in scheduled or repeatable jobs, like data processing pipelines or rendering tasks.
- Machine learning and AI: Model training and experimentation that can be checkpointed and resumed as needed.
- Scaling compute-heavy workloads: When used alongside on-demand resources, spot VMs can help increase capacity at a lower cost.