Renting & Deploying a GPU

GPUnex is not only a platform for GPU package holders — it is also a full-featured GPU compute marketplace for AI engineers, researchers, and companies that need on-demand access to enterprise-grade hardware. This guide walks you through the entire process: applying for renter access, selecting a GPU model, choosing a pricing plan, deploying your instance, and connecting to it.

Applying for Renter Access

GPU rental access requires an approved application. This verification step ensures quality of service for all marketplace participants and helps GPUnex match you with the right hardware for your workload.

Step-by-Step Instructions

1. Navigate to the Renter section by clicking “Rent GPUs” in the sidebar, or go to the Roles page and select the Renter role.

2. Click “Apply to Rent GPUs” to open the application form.

3. Fill out the application with the following details:

   Personal / Company Information:

   • Full name (required)
   • Email address (required)
   • Company name (optional — individuals are welcome to apply)

   Use Case Details:

   • Use case description — Describe what you plan to use the GPU compute for (e.g., training a large language model, running inference workloads, rendering, scientific simulation)

   GPU Requirements:

   • Preferred GPU model — Select from available models (H100, A100, L40S, L4)
   • Quantity — How many GPUs you need (single GPU or multi-GPU configurations)
   • VRAM requirements — Minimum VRAM needed for your workload
   • Usage pattern — Continuous, burst, or periodic usage
   • Preferred region — Select your preferred EU data center region for lowest latency

   Budget and Timeline:

   • Estimated monthly budget in USDC
   • Timeline — When you need to start (immediately, within a week, within a month)

4. Review your application for accuracy and completeness.

5. Click “Submit” to send your application for review.

Application Review

• Applications are reviewed within 24-48 hours during business days.
• You will receive an email notification when your application is approved.
• Once approved, you gain full access to the GPU rental marketplace and can deploy instances immediately.

Available GPU Models

GPUnex offers a range of NVIDIA data center GPUs to match different workload requirements and budgets.

Multi-GPU Configurations

For workloads that require more compute than a single GPU can provide, GPUnex supports multi-GPU configurations:

• Up to 8x NVIDIA H100 GPUs in a single node
• NVLink interconnect for high-bandwidth GPU-to-GPU communication within a node
• InfiniBand networking for multi-node clusters
• Ideal for distributed training of large models, large-scale inference serving, and research requiring massive parallelism

Choosing the Right GPU

Use the following guidelines to select the appropriate hardware:

• Large language model training (>7B parameters): H100 80GB or A100 80GB, multi-GPU recommended
• Fine-tuning pre-trained models: A100 40GB or A100 80GB depending on model size
• Production inference serving: L40S 48GB or L4 24GB for cost-efficient inference; H100 for latency-critical applications
• Rendering and visualization: L40S 48GB offers excellent graphics and compute performance
• Batch processing and lightweight inference: L4 24GB provides the best cost-per-inference for smaller models

Pricing Models

GPUnex offers three pricing models to fit different usage patterns and budget requirements.

On-Demand

• Billing: Per-second, pay only for what you use
• Commitment: None — start and stop at any time
• Flexibility: Full flexibility to scale up or down instantly
• Best for: Experimentation, variable workloads, short-term projects

Reserved

• Commitment: Days, weeks, or months
• Discount: Up to 40% off On-Demand pricing
• Capacity: Guaranteed — your reserved GPU is always available during your commitment period
• Best for: Ongoing training runs, production inference, predictable workloads

Spot

• Discount: Up to 70% off On-Demand pricing
• Preemption: Instances may be reclaimed with a 30-second warning when demand spikes
• Auto-checkpoint: The system automatically triggers a checkpoint save when a preemption notice is issued, minimizing lost work
• Best for: Fault-tolerant workloads, batch processing, non-time-critical training jobs, budget-conscious experimentation

Deploying an Instance

Once your renter application is approved, you can deploy GPU instances from the marketplace.

Step-by-Step Instructions

1. Navigate to the GPU Marketplace from your dashboard.

2. Search and filter the available inventory using the following criteria:

   • GPU model (H100, A100, L40S, L4)
   • VRAM (24GB, 40GB, 48GB, 80GB)
   • Region (select your preferred EU data center location)
   • Price range (filter by hourly rate)
   • Pricing model (On-Demand, Reserved, or Spot)

3. Select a GPU from the search results. Click on the listing to view full details including specifications, pricing, availability, and data center location.

4. Choose your software environment. Select from pre-configured frameworks or provide your own:

   Pre-installed frameworks:

   • PyTorch — Latest stable release with CUDA support
   • TensorFlow — Latest stable release with GPU acceleration
   • JAX — High-performance ML framework
   • Jupyter — JupyterLab environment pre-configured with common ML libraries

   Custom environment:

   • Provide a custom Docker image from any public or private container registry
   • Specify the image URL and any required environment variables

5. Configure instance settings:

   • Instance name (for your reference)
   • SSH public key (for secure access)
   • Startup script (optional — runs automatically when the instance launches)
   • Environment variables (optional)

6. Review and deploy. Confirm the GPU selection, pricing model, software environment, and configuration. Click “Deploy” to launch your instance.

7. Wait for provisioning. Your container will be deployed and an SSH connection will be available within seconds of clicking Deploy.

Connecting via SSH

Once your instance is running, you will receive connection details on the instance management page:

• The IP address, port, and username are provided in the instance details panel.
• Use the private key corresponding to the SSH public key you provided during deployment.
• You will have root access inside the container.

Managing Your Instance

From the instance management page, you can:

• Start / Stop the instance to pause billing (On-Demand only)
• Monitor GPU utilization, memory usage, and compute metrics in real time
• View logs from your running processes
• Terminate the instance when you are finished

Persistent Storage

GPUnex provides network-attached persistent storage that is independent of your GPU instance lifecycle.

Key Features

• Survives restarts. Your data remains intact when you stop and restart an instance.
• Attachable to new instances. If you terminate one instance and deploy a new one, you can attach your existing storage volume to the new instance. All your data, model checkpoints, datasets, and code are preserved.
• Data persistence. Your storage volume persists until you explicitly delete it, regardless of what happens to your compute instances.

Use Cases for Persistent Storage

• Store training datasets that you use across multiple training runs
• Save model checkpoints so you can resume training after interruptions (especially important for Spot instances)
• Maintain code repositories and development environments across instance changes
• Keep experiment logs and results for long-running research projects

Getting Help

Standard Support

All renters have access to the GPUnex support system for technical issues, billing questions, and general assistance. Reach out through the support section in your dashboard.

Enterprise Support

Enterprise customers and teams with larger deployments receive additional support:

• Dedicated account management with a named point of contact
• Priority support with faster response times
• Custom cluster configurations tailored to your specific workload requirements
• Volume pricing for large-scale deployments
• SLA guarantees for uptime and performance

Need a Custom Solution?

For custom cluster configurations, multi-node deployments, reserved capacity agreements, or enterprise pricing, our sales team is ready to help. Whether you need a dedicated H100 cluster or a tailored infrastructure setup, we can build the right solution for your workload.

Contact Sales to discuss your requirements and get a personalized quote.

Next Steps

• API Reference — Integrate GPUnex GPU management into your own scripts and automation pipelines
• How GPU Packages Work — Learn about the investor side of the GPUnex marketplace
• Contact Sales — Get in touch for enterprise solutions, custom configurations, and volume pricing