GPU Server Blibs

1. Access and Overview

After logging in at https://james.trooper.ai/manage , users are presented with a list of active and frozen Blibs. Each server entry includes the following information:

• Blib name and status (running, stopped, frozen), read more: Blibs Status List
• GPU configuration (e.g. 1× RTX 3090, 2× A100)
• CPU and memory (e.g. 12 cores, 86 GB RAM)
• NVMe SSD capacity
• Assigned hostname (e.g. connect01.trooper.ai)
• Allocated public port range (TCP, optional UDP)
• Installed software Templates (e.g. Jupyter Notebook, ComfyUI Template for GPU Server Blibs)
• Secure HTTPS access, read more: 🌍 HTTPS Access (SSL Proxy)
• GPU Drivers
• Backups and Snapshots

System temperature and billing periods are also shown where applicable.

2. Creating and Starting a Server

To deploy a new GPU Server Blib:

1. Navigate to the Order GPU Server section or use the Create New Server function in the dashboard.
2. Select the desired server configuration (e.g. sparbox.m1, infinityai.m2).
3. Optionally, choose one or more templates to be pre-installed:
   • Jupyter Notebook
   • OpenWebUI & Ollama
   • ComfyUI Template for GPU Server Blibs
4. Confirm the selection and accept the terms.
5. Start the server.

Provisioning usually takes 1-5 minutes, depending on the server you choose. After that, the system will install drivers and monitoring tools, which can take another 2-10 minutes depending on external APT mirror speed. You can expect one automatic restart during this process as the GPU drivers are set up.

🚫 Server Management Software: Do not install a server management software like Dokploy, CapRover, Plesk, etc. This will not work on a GPU server. These tools are made for your Web Server and than connecting the apps to your GPU server via API.

🚫 Blackwell Driver Downgrades: As per our terms of service, do not modify the drivers for Blackwell cards (50xx, RTX Pro xx00) on your machine BELOW version 580-server-open. Doing a driver downgrade may cause system instability and potential service interruption. Your server will remain stopped until the issue is resolved. No refund. CUDA’s backwards compatibility eliminates the need for older driver versions.

Current prices of our GPU servers:

** Promotional pricing reflects the maximum discount available through prepaid budget options. Currently +50% Extra Credits! See the “Add Budget” function for details.

Order now

3. Stopping, Freezing, Resetting and Terminating GPU Server Blibs

Trooper.AI (Service Introduction) provides flexible lifecycle options for each Blib:

Productive states:

• Start: Starts the server.
• Stop: Halts the server. Resources (GPUs, CPU Cores, RAM) remain allocated, and the instance can be restarted within seconds.

Optional states:

• Freeze: The server is archived and resource allocation is released. Startup time may vary and reassignment of ports may be required. Consider using the stopped state if predictable startup times are needed! FREEZING IS NOT RECOMMENDED FOR ENTERPRISE CLIENTS DUE TO PORT CHANGES AND LONGER TIMES ON INACCESSIBILITY!
• Reset: Deletes all data on the server and reinstalls the initial configuration. Ports may change. CPU model and speed may change. Your pricing stays. ALL DATA WILL BE LOST! This action cannot be undone.
• Terminate: Permanently deletes the server and all associated data. This action cannot be undone.

Read more: Blibs Status List

All lifecycle actions are accessible via the context menu in the dashboard.

4. Template Management

Software templates can be selected during server creation or installed afterwards. Templates automatically install dependencies, configure services, and open required ports.

Common templates include:

• Jupyter Notebook
• OpenWebUI & Ollama
• ComfyUI Template for GPU Server Blibs
• Ubuntu Desktop and RDP Access
• Framepack
• Any Docker
• … and many more Templates

Multiple templates can be combined as needed on one server blib.

5. Public Network Access

Each Blib is assigned a dedicated range of public TCP ports (typically 10–40 open 1:1 NAT ports). These are accessible via the assigned hostname (e.g. connectXX.trooper.ai) and are displayed in the dashboard.

Connection Limits (Outbound):
To ensure platform stability and prevent abuse or misconfigured workloads from generating excessive external traffic, all Blibs are subject to per-machine outbound rate limits and connection tracking limits. This includes:

• Maximum number of concurrent outbound connections (2048 per GPU)
• TCP SYN rate limiting (new outbound connections)
• UDP packet rate limiting (NEW state)
• ICMP rate limiting

If your machine exceeds these thresholds, additional outbound traffic may be temporarily dropped or rejected by the routing layer.

Timeout Handling:
For security reasons, long-lived connections may be periodically closed by our routing system. We recommend implementing robust reconnection and timeout handling within your software to ensure continued operation.

UDP Traffic (Optional):
UDP traffic is not enabled by default. You can enable UDP access for your Blib via the Dashboard if required. Please ensure your application is designed to safely handle UDP exposure.

For detailed information on accessing your Blib, please refer to the following documentation:

• SSH Access (incl. File Transfer)
• RDP Access
  • optional: Realtime Access with NoMachine
• 🌍 HTTPS Access (SSL Proxy)
• About SSL Certificates
• Do I need a Firewall on the GPU Server?

6. Monitoring and System Information

The management interface provides the following real-time monitoring features:

• CPU and GPU temperature display
• Network traffic information
• Resource utilization (CPU, RAM, GPU)
• Template status and version
• Uptime and lifecycle state

7. Behavior on Insufficient Credit

Blibs operate on a prepaid credit system. If the prepaid balance is depleted:

• The server is automatically stopped and may be frozen.
• If the balance is not replenished within 72 hours, the server and all associated data may be permanently deleted.

Note: Servers will not accrue charges without available credit. However, GPU resources are not guaranteed unless the balance is sufficient.

8. Contract Models and Resource Use

• All GPU servers include 5 TB of monthly traffic per GPU. For example, a 2-GPU server includes 10 TB.
• If usage exceeds this soft limit, bandwidth may be throttled or additional charges applied.
• Discounts are available for weekly, monthly, and annual contracts.
• Yearly contracts and business invoicing options are available upon request.

10. Support

For technical issues or custom provisioning requests, please contact:

• Email: support@trooper.ai
• WhatsApp: +49 6126 9289991

11. Migration / Upgrade of Blibs

Upgrade your existing GPU Server to a more powerful configuration is a great thing on Trooper.AI and very unique to us. This gives you the ability to seamlessly transition to more powerful hardware without losing your existing setup or data.

During migration, your server will be temporarily frozen. You’ll only see configurations with better or equal resources available. A migration service fee plus addition remaining interval costs, as displayed during the process, may apply.

Ports may change.

Just select “Migrate to better Blib” from the Actions pulldown on your server.

12. Bare Metal and Virtualization Architecture

Blibs utilize a hybrid architecture, combining dedicated bare-metal resources with virtualized components:

This approach delivers 100% of the performance potential of the GPU, CPU, and RAM, while leveraging the flexibility of virtualized NVMe storage and networking for streamlined provisioning, robust backups, and enhanced security.

You also benefit from full control over NVIDIA drivers, kernel modules, and APT package handling. This makes your GPU Server Blib a reliable and scalable AI machine for daily use and production environments.