Compute Modes

Choose the best compute strategy: priority, commit or flex — depending on your needs.

Choose How You Use the Grid. Precision-Priced Compute for Every Scenario.

Skiv AI offers multiple compute execution modes so you can match your workload’s urgency, stability requirements and budget — without sacrificing access to performance-grade GPU infrastructure.

🚨 Priority Mode (Guaranteed Runtime)

• Full Resource Lock You get exclusive access to a GPU or multi-GPU node. No job interruptions. No capacity sharing.

• Ideal for Critical Tasks Recommended for model training, production inference or time-sensitive research that must complete without fail.

• Predictable Runtime Book by the hour or for fixed durations. Jobs run uninterrupted until completed or stopped by the user.

• Auto-Renew Optional Extend session duration automatically if capacity is available or manually request continuation before expiry.

💸 Commit Mode (Discounted Reservations)

• Prepay, Save More Lock in compute resources ahead of time and receive up to 50% off standard rates depending on duration and node type.

• Flexible Withdrawal Terms Early exits possible with partial refund depending on usage and uptime. Works best for repeat jobs with consistent scheduling.

• Balance of Commitment & Cost A middle ground — great for teams looking to stretch budget without sacrificing performance.

🧪 Flex Mode (Interruptible Compute)

• Dynamic Pricing via Bidding Submit a bid for GPU usage. Higher bids get priority, lower bids may experience pause or preemption during high demand.

• Best for Non-Critical Jobs Great for render queues, dataset pre-processing or batch jobs that can restart without issue.

• Save Up to 70% Maximize cost efficiency with a flexible schedule — but implement checkpointing to prevent data loss.

• Built-in Monitoring & Retry Flex jobs auto-resume when capacity is available. Retry logic built into the task scheduler.