GPU Instances

GPU Instances are full virtual machines with dedicated GPU hardware, persistent storage, and complete root access. They’re ideal for training, fine-tuning, and development workflows.

Available GPU Types

GPU	VRAM	Use Case	Price/hr
NVIDIA H100 SXM	80 GB	Large model training	$3.89
NVIDIA A100 SXM	80 GB	Production training	$1.64
NVIDIA RTX 4090	24 GB	Fine-tuning & inference	$0.44

Cloud Types

• Secure Cloud — Enterprise-grade data centers with SOC2 compliance, redundant power, and guaranteed uptime SLA.
• Community Cloud — Cost-effective compute from verified providers. Lower price, slightly less reliability guarantees.

Instance Lifecycle

CREATED → STARTING → RUNNING → STOPPING → STOPPED → TERMINATED

• RUNNING — Instance is active and accessible via SSH/web terminal
• STOPPED — Instance is stopped, storage is preserved, no GPU charges
• TERMINATED — Instance is deleted, all data is lost unless using network volumes

Connecting to Your Instance

SSH Access

ssh root@<instance-ip> -p <ssh-port>

Your SSH port and IP are shown in the dashboard once the instance reaches RUNNING status.

Jupyter Lab

Many templates include JupyterLab. Access it at:

https://<instance-id>-8888.proxy.gpuworker.com

Persistent Storage

Each instance has:

• Container Disk — Fast local SSD storage, lost on termination
• Volume Mount — Optional network volume at /workspace, persists across instance restarts

For data that must survive instance termination, use Storage Volumes.

Best Practices

1. Use templates — Save your environment as a template to speed up future deployments
2. Stop when idle — Stop instances when not in use to avoid GPU charges (storage charges still apply)
3. Use network volumes — Store datasets and checkpoints on network volumes for persistence
4. Monitor costs — Check the billing dashboard regularly to track spending