Use your GPU

Storage

/workspace and shared volumes.

Each GPU has two kinds of storage.

/workspace

Every GPU has its own disk at /workspace.

  • Kept when you pause and resume, and when the GPU restarts.
  • Deleted when you terminate the GPU.

Use it for code, scratch data and anything you can recreate.

Shared volumes

A shared volume is a disk that belongs to your team, not to one GPU.

  • Mounted at /mnt/<volume-name> in any GPU you attach it to.
  • Kept when GPUs are terminated. Delete it yourself from Storage when you no longer need it.
  • Several GPUs on your team can mount the same volume at once, so one GPU can write checkpoints another reads.
  • Billed at US$0.10 per GB-month while it exists, attached or not.

Create one

  • When launching: in the launcher's Storage step choose New volume and a size, or Existing to attach one you have.
  • Later: open a running GPU, then More → Attach storage. Attaching to a running GPU restarts it so the volume can be mounted; /workspace is kept.

You can attach up to 3 volumes to a GPU.

Manage volumes

The Storage page lists your volumes, their size and whether they're in use. Delete a volume from there once no GPU is using it. Deleting a volume deletes its data; this can't be undone.

Which should I use?

DataPut it in
Datasets, model weights, checkpoints you want to keepA shared volume
Code you also keep in git, temporary files/workspace
Results you need after terminatingA shared volume, or copy them off first

Flint documentation