DGX Spark pricing. From $0.79 an hour.

Per-minute, against a prepaid credit balance. Running instances bill the full hourly rate; a customer-requested stop keeps its Spark held for you at a 25% discount, so $0.59/hour while stopped, against $0.79/hour running. Top up $10 the first time and get $20 in credit (a dollar-for-dollar match of up to $10 on your first top-up); after that, the minimum top-up is $10. If your balance runs out, your instances are stopped and their machines released, and each workspace is saved for 7 days before it is deleted.

No. The hourly rate is the whole price, ingress and egress are included. (The 200 Gb/s ConnectX-7 ports on the Spark are a node-to-node interconnect, not your internet uplink.)

Stopping keeps /workspace right where it is, on the same node: the Spark stays held for you, the container stops, and the SSH port and hostname do not change, so restarting keeps the same machine. Model presets may need several minutes to warm up. A customer-requested stop bills 75% of the hourly rate to keep the Spark held, a 25% discount. To stop paying, terminate the instance instead of stopping it.

Terminating deletes your workspace from the node for good. That is why it asks you to type the instance name to confirm.

If your credit reaches zero or you hit your monthly budget, the instance is stopped and the machine is released for someone else to rent. Your workspace is saved off it and kept for 7 days: top up and press start within that window to pick up where you left off, on whichever Spark is free. After 7 days without that, the saved workspace is deleted for good. If the copy cannot be made, the machine stays held for you for up to 24 hours, then it is wiped with no copy kept. Low-balance mails warn you before this happens, and auto top-up can reduce the risk, but payments may fail. Keep your own backup regardless.

Workspaces are not backed up. While an instance runs or is stopped, GPUwerk copies its /workspace to one of its own machines on the same network about every six hours, only so it can recover from a failed drive. That copy can be hours old or catch a file mid-write, and nothing guarantees it exists or is complete. Keep your own copy of anything you need.

A GB10 machine for your own site is quoted in writing, sized to your workload; any help with setup or operation is scoped on request. Write to us with your country and use case and you'll get a written quote, not a sales call. To own a GB10 without running it yourself, buy one in the GB10 store and have GPUwerk host it.

You join a first-in, first-out line for that tier. Joining is free and needs no credit; if no DGX Spark is free when you create your account, we add you to the Spark line automatically. Machines are not held for anyone. When one of that tier frees up, we email people in line one at a time, longest-waiting first, a few minutes apart, and whoever deploys first from the console gets it. Each place in line gets that email once; if the machine is gone when you look, you keep your place. The two-Spark cluster has no emails: deploy it from the console when two Sparks show as free.