GPU Pricing
Simple, transparent pricing billed by the second. No hidden fees, no minimum commitments. Pay only for what you use.
RTX 4090
Ada Lovelace Architecture
$0.00010/sec per GPU
RTX 5090
Blackwell Architecture
$0.00013/sec per GPU
RTX PRO 6000
Blackwell Architecture
$0.00022/sec per GPU
Infrastructure Costs
Additional costs for storage and network bandwidth, billed separately from compute.
Storage
NVMe local storage. Pay only for the storage you provision.
Network
Network data transfer billed per TB. Up to 5Gbps network transfer bandwidth.
How Billing Works
Simple, transparent billing with no surprises.
Per-Second Billing
You're charged by the second, not the hour. Stop your instance at any time and only pay for the exact seconds you used.
Interruptible Instances
Get up to 80% off with interruptible instances. Your instance may be interrupted by on-demand or reserved instances, ideal for fault-tolerant workloads.
Reserved Pricing
Commit to 3 or 6 months for 10-20% off on-demand rates. You must pre-pay the full amount of the GPU rental cost for the rental period to create a reserved instance.
Frequently Asked Questions
How does per-second billing work?
Your GPU usage is tracked to the second. When you stop or delete your instance, billing stops immediately. There are no minimum usage periods or rounding to the nearest hour.
What's the difference between on-demand and interruptible?
On-demand instances run until you stop them. Interruptible instances cost significantly less but may be interrupted by on-demand or reserved instances. They're ideal for training jobs with checkpointing.
How do reserved instances work?
Reserved instances work the same as on-demand, but you commit to a fixed period (3 or 6 months) and receive a 10-20% discount in return. The GPUs are reserved for you for the full term. Ideal if you need a longer-running instance.
Ready to get started?
Deploy a GPU instance in minutes. New users get $1 free credit to try instantly.