GPU Accelerated Compute
Hubfly space GPU Deployments let you provision on-demand accelerated compute for artificial intelligence, machine learning, model inference, PyTorch, CUDA workloads, and Jupyter notebook experimentation.
Dedicated GPU Wallet & Offers
GPU Deployments operate on live rentable GPU offers with hourly billing managed through a dedicated GPU Wallet. Ensure your GPU wallet maintains a positive balance before provisioning.
GPU Instance Sizing & Features
| Specification | Description & Parameters |
|---|---|
| Hardware Offers | Search verified NVIDIA GPU offers filtered by GPU model, VRAM capacity, TFLOPS, CPU cores, host RAM, and hourly price. |
| Disk Allocation | Specify disk size per GPU instance (minimum 10 GB up to offer maximum) to accommodate model weights, datasets, and package caches. |
| GPU Templates | Prebuilt container environments featuring CUDA drivers, PyTorch, TensorFlow, SSH direct access, and automatic Jupyter notebook endpoints. |
| Binding Scope | Project-Bound (tied to a specific project workflow & topology) or Global (shared account research capacity). |
| Data Copying | Direct GPU-to-GPU data transfer utility to copy workspace files between instances before replacing hardware. |
Accessing Jupyter Notebooks
When deploying a Jupyter-enabled template, Hubfly space automatically extracts the public IP, mapped port, and security token to build your one-click Jupyter endpoint URL.