GPU Blades
SSH into real GPUs in minutes. Multi-user isolation, auto-provisioning, and pre-configured environments for AI competitions, workshops, and courses.
Three Steps to GPU Access
Install
pipx install rb-cli
Works on macOS, Linux, and Windows/WSL.
Join
rb event join CUDA-2026
Enter the event code shared by your instructor.
Connect
rb blade ssh
SSH into your GPU. PyTorch, CUDA, and tools pre-installed.
Available GPUs
From lightweight demos to multi-GPU distributed training.
| GPU Type | VRAM | Best For | Price |
|---|---|---|---|
| RTX 4000 Ada | 20 GB | Light workloads & demos | €0.99/hr |
| L40S | 48 GB | Training & workshops | €1.99/hr |
| RTX 6000 Ada | 48 GB | Training & workshops | €1.99/hr |
| MI300X | 192 GB | Large model inference | €2.49/hr |
| H100 | 80 GB | Large model training | €4.29/hr |
| H200 | 141 GB | Large model training | €4.49/hr |
| L40S x4 | 192 GB | Large model training | €7.99/hr |
| L40S x8 | 384 GB | Distributed training | €15.99/hr |
| MI300X x8 | 1536 GB | Multi-GPU inference | €19.99/hr |
| H100 x8 | 640 GB | Multi-GPU workloads | €29.99/hr |
| H200 x8 | 1128 GB | Multi-GPU workloads | €34.99/hr |
GPU Availability
Live availability status across our GPU fleet.
Built for GPU Workloads
Multi-User Isolation
Up to 20 users share a single GPU with full container isolation. Each user gets their own environment.
Auto-Provision at Event Start
GPU blades spin up automatically when your event starts. No manual provisioning needed.
Auto-Teardown at End
When the event finishes, blades are destroyed automatically. No forgotten instances running up bills.
Pre-Configured Environments
PyTorch, TensorFlow, JAX, and CUDA pre-installed. Start training immediately.
VS Code Remote-SSH
Connect with VS Code Remote-SSH for a full IDE experience on GPU hardware.
Blade Failure Auto-Recovery
If a blade fails, a replacement is provisioned automatically. Your users stay connected.
Everything From the Terminal
The rb CLI gives you full control over your GPU blades.
Ready to SSH into a GPU?
Install the CLI and connect to a GPU blade in minutes.