Create an Instance

Use this sequence to choose hardware, configure access, and confirm the displayed price before an instance is created.

Before you start

• Sign in to the ANK dashboard.
• Add wallet balance for a paid configuration. Usage is billed in USD; you can top up in USD or NPR.
• Keep an SSH public key ready if you want key-based access. You can also use a username and password.

Select the workload

The dashboard shows each tier's hardware, CPU, RAM, disk, and hourly rate before you create an instance.

• NVIDIA GeForce RTX 3090 (S, M, 2X), NVIDIA GeForce RTX 5080 (S, M, L), and NVIDIA GeForce RTX 5090 (S, M, XL, XXL, 2X) are the currently provisionable GPU tiers. The RTX 5080 S configuration, for example, includes one GPU, 8 CPU cores, 16 GB RAM, 100 GB disk, and starts at NPR 52.92/hour.
  • 2X tiers are two independent GPU instances on two separate hosts rather than one bigger machine — read Multi-Node Instances before requesting one.
  • Other GPU options (RTX PRO 6000, DGX Spark) are listed for capacity discussions. Confirm availability with ANK before planning a workload around them.

Choose a billing plan that fits the run. Pay As You Go is billed hourly; daily, weekly, and eligible monthly plans apply longer-term discounts. See Pricing and Billing for the current plan details.

Configure the instance

1. Click New Instance.
2. Select the tier, configuration, and billing plan.
3. Select an image:
   • ubuntu-cpu — Ubuntu 26.04 for CPU-only work.
   • ubuntu-gpu — Ubuntu 26.04 with CUDA 12.8 drivers pre-installed.
   • ubuntu-torch — Ubuntu 26.04 with PyTorch 2.8 and CUDA 12.8 preloaded, for GPU-ready PyTorch workloads.
4. Set a username and password, add one or more SSH public keys, or use both. Add public keys only; keep the matching private keys on your own device.

Review, create, and verify

Before you create the instance, confirm the selected configuration, image, access method, displayed hourly rate, and wallet balance. Then create the instance.

Provisioning status appears in the dashboard. Wait until the instance is active before connecting.

Once it is active, continue to Connect and Verify. Run the connection and GPU checks there before starting a long workload.

Before committing to a long run

Use a short representative job to confirm the image, access method, displayed rate, and available capacity. Keep the tier, configuration, image, billing plan, and setup steps with your project notes so a successful run is easy to repeat.