​
Overview

Use this section to understand some of the core concepts we use at GPU Trader. You will find links to more detailed documentation in these sections, and as always, reach out to us for help.

​
Concepts

​
Access Top-Tier GPUs

GPU Trader gives users instant access to enterprise-grade GPUs located in professionally managed data centers around the world. Whether you’re fine-tuning a model, running multi-GPU training, or powering inference at scale, our marketplace connects you to powerful hardware — no long-term contracts, no hardware management, and no waiting in queue. Provision what you need, when you need it through our console or API.

​
Supported GPU Models

We support a growing catalog of NVIDIA’s most capable data center GPUs. This includes the latest H100 and H200 cards for state-of-the-art AI workloads, A100s for balanced performance and cost, and specialized models like the GH200 for memory-intensive tasks. You can filter by architecture, VRAM, interconnect type (PCIe or SXM), and more—ensuring the hardware matches the needs of your workloads.

​
On-Demand and Reserved Instances

GPU Trader offers two flexible modes of access. On-demand lets you launch instances immediately with no commitments, ideal for burst workloads or experimentation. Reserved instances allow you to secure GPU capacity for a set period at a fixed price, which is perfect for longer training runs, scheduled jobs, or production deployments. Both modes use the same provisioning flow, so you can switch between them as your needs evolve.

​
Easy Setup and Management

Deploying on GPU Trader is fast and frictionless. Use our browser-based console, or API to launch instances within minutes. Choose the number of cards for your instance. Then boot with a clean, container-friendly Linux environment while you retain full control. Use the web terminal, secure tunnels, or SSH to connect to your instance. Job monitoring, GPU metrics, and billing visibility are all built into your dashboard, giving you a seamless way to manage infrastructure without being an infrastructure company.

​
Pre-Configured Templates

Don’t waste time installing dependencies. GPU Trader offers pre-built templates for common AI and ML use cases—from fine-tuning LLMs with PyTorch to serving models with vLLM or NVIDIA Triton. Each template is a tested container image with the right drivers, runtimes, and libraries, so you can get to work without wrestling with CUDA versions or compatibility issues.

​
Transparent Billing and Pricing

You pay only for what you use—at the card level. Each GPU instance is priced by the hour, per card. We meter at 1-minute intervals to help you control costs. Your dashboard provides usage logs, plus payout transparency if you’re a provider. Whether you’re renting capacity or monetizing idle GPUs, you’ll always know where the money’s going.