Rig Launch · GPU
GPU servers. By the hour.
NVIDIA L4 GPUs with 24 GB of memory each, on their own hosts, with the driver and CUDA already installed. Start one for a job and delete it when the job is done.
Sizes · USD
Three sizes
| Size | GPUs | Server | Price | |
|---|---|---|---|---|
| GPU Small | 1 × NVIDIA L4, 24 GB each | 4 vCPU · 16 GB memory · 100 GB disk | … | Choose GPU Small |
| GPU Medium | 1 × NVIDIA L4, 24 GB each | 16 vCPU · 64 GB memory · 200 GB disk | … | Choose GPU Medium |
| GPU Large | 4 × NVIDIA L4, 24 GB each | 48 vCPU · 192 GB memory · 400 GB disk | … | Choose GPU Large |
What it's for
Jobs that need a GPU for a while
Models
Run an open model behind your app, or try one before you commit to it. An L4's 24 GB holds most small and mid-size models.
Media and batch jobs
Video transcoding, image processing, embeddings for search: start a server, run the job, delete it.
How it works
What to know
Billing and capacity
GPU servers are billed by the hour only, from when they are ready until you delete them, and never more than the monthly price shown.
- Ubuntu with the NVIDIA driver, CUDA and the container toolkit installed
- The same panel, firewall, backups and SSH as every Launch server
- If a region is full you join a waitlist and nothing is charged; when a place frees up it is held for you for 48 hours
Also in Launch
Keep going
- Overview Plans, prices and how a deploy goes.
- The Duke Rig's AI agent: deploys, code review, failures and setup, each step after you approve it.
- Servers Linux servers: SSH, networking, backups, resizing, monitoring and support access.
- Apps Ship from Rig or Git without a server, or run managed Kubernetes.
- Databases Managed Postgres and Redis, started from the database already on your Mac.
- Storage Extra disks, S3-compatible buckets with a CDN, your own images, and free DNS.