Cost Tracking
Recaster tracks your cloud GPU spending in real time and lets you set budget limits with automatic alerts. Stay in control of Vast.ai costs without surprises.
Overview
Cloud GPU instances on Vast.ai are billed by the hour. Costs depend on the GPU model you select, with prices typically ranging from $0.10 to $1.50 per hour. While this is affordable for short sessions, costs can add up during long training runs or if you forget to stop an instance. The cost tracking system gives you visibility into your spending and tools to prevent overspending.
Vast.ai handles billing directly
Budget Configuration
Set daily and monthly spending limits to control your cloud GPU costs. When your spending approaches or reaches a limit, Recaster displays warnings and can optionally stop running instances automatically.
| Setting | Range | Default | Description |
|---|---|---|---|
| Daily Limit | $0 – $100 | None (unlimited) | Maximum spending per calendar day |
| Monthly Limit | $0 – $1,000 | None (unlimited) | Maximum spending per calendar month |
| Warning Threshold | 50% – 95% | 80% | Percentage of budget that triggers a warning notification |
| Auto-Stop | On / Off | Off | Automatically stop all instances when budget is reached |
To configure budget settings, open the Budget Configuration dialog from the Remote panel toolbar. Set your limits, choose a warning threshold, and optionally enable auto-stop. Settings are saved per-project.
Auto-stop may interrupt training
Per-Session Costs
Recaster tracks the cost of each cloud session individually. You can see how much each training run, Quick Recast job, or upscaling session has cost. This information is available in the Remote panel and the Multi-Session dashboard.
Session Cost Breakdown
- GPU rental: Per-hour rate from Vast.ai multiplied by session duration
- Storage: Disk usage charges (typically much smaller than GPU cost)
- Network: Data transfer for file uploads and downloads (usually minimal)
- Total: Sum of all cost components for the session
Session costs are updated in real time while the instance is running. After the session ends, the final cost is recorded in the training history for future reference.
Warnings and Alerts
When your spending approaches the configured budget limit, Recaster provides graduated warnings:
A yellow notification appears in the Remote panel and the status bar. You can continue running instances, but you are approaching your limit.
An orange alert appears prominently. Consider stopping instances or saving your work soon.
A red alert indicates the budget limit has been reached. If auto-stop is enabled, all instances are stopped. If auto-stop is off, you can still run instances but the warning persists.
Multi-Session Tracking
When running multiple concurrent sessions (for example, training on one instance while upscaling on another), the cost tracker aggregates spending across all active instances. The Multi-Session dashboard in the Remote panel shows:
- Individual cost per session with instance type and GPU model
- Total combined cost for all active sessions
- Daily and monthly running totals
- Projected cost based on current run rates
Optimization Tips
Follow these tips to minimize your cloud GPU spending while maintaining productivity:
Choose the right GPU
Not every task needs a top-tier GPU. Quick Recast processing and video upscaling work well on mid-range GPUs like the RTX 3080 at a fraction of the cost. Reserve high-end GPUs (RTX 4090, A100) for long training sessions where faster processing saves enough time to justify the higher hourly rate.
Stop instances when not in use
This is the single most impactful cost-saving measure. Cloud instances charge by the hour whether or not you are actively processing. If you step away for lunch or end your work session, stop the instance from the Remote panel. Restarting later takes only 2 to 5 minutes.
Pause training instead of running overnight
Unless you specifically need overnight training, pause training and stop the instance before leaving. An 8-hour overnight session on an RTX 4090 at $1.00/hr costs $8.00, which may produce only marginal improvement over shorter, focused sessions.
Batch your uploads
Prepare all your source material before launching an instance. Upload everything at once, then run your processing jobs. This avoids idle time while the instance is running but you are still preparing files locally.
Use presets for upscaling
Run a quick test with the Fast preset on a short clip before committing to a full-length upscale with the Quality preset. This lets you verify settings without spending hours on the wrong configuration.
Monitor spending regularly
Check the cost display in the Remote panel periodically. Set a reasonable daily budget limit and enable the warning threshold so you are alerted before reaching your limit.
Recommended budget for getting started
Cost Examples
The following examples illustrate typical costs for common workflows. Actual prices depend on Vast.ai availability and the specific GPU you select.
| Task | GPU | Duration | Estimated Cost |
|---|---|---|---|
| Quick Recast (2 min video) | RTX 3080 | ~15 min | $0.05 – $0.10 |
| Upscale 4x (2 min video) | RTX 3090 | ~10 min | $0.05 – $0.15 |
| SAEHD training (2 hours) | RTX 4090 | 2 hours | $1.40 – $3.00 |
| Full-day training | RTX 3090 | 8 hours | $2.40 – $5.60 |
Cost History
All session costs are recorded in the training history alongside other session metadata. You can review past spending in the training history panel, which shows the cost for each completed session along with the GPU model, duration, and instance type used.
This historical data helps you understand your spending patterns over time and make informed decisions about GPU selection and session length.
Budget resets
Was this page helpful?