Skip to main content

Features

What a Block Storage disk can do, and the limits that come with it.

Read-Write Mounts​

A disk mounts at /datasets/<name> with write access. This is the main reason to choose it over Object Storage, which mounts read-only: anything that downloads, unpacks, transforms, or checkpoints into its mount directory needs a disk.

A disk attaches to one resource at a time. If it is already mounted elsewhere, the portal reports it as already mounted rather than attaching it twice.

Low-Latency Access​

The disk is attached directly to the workload it is mounted on, so random reads and writes are fast and local. That makes it the right place for:

  • Scratch space during data preparation.
  • A cache in front of a larger Object Storage bucket.
  • Checkpoints written frequently during long training runs.

For data that must be read concurrently by many nodes, use SFS or PFS instead.

Resizable Capacity​

A disk is created at a fixed size — minimum 10 GB, 100 GB by default — and can be increased at any time from the Resize Disk tab, up to your project's storage limit.

Growth is one-way

A disk can never be reduced. You are billed on the allocated size regardless of how much you use, so over-provisioning costs money you cannot recover without creating a smaller disk and migrating.

If you hit the project ceiling, the resize panel offers a Request button that submits a limit increase for review. See Grow the Disk Later.

Encryption​

Block Storage supports encryption at rest with E2E Managed keys. Turn it on with the Enable Encryption switch when creating the disk; user-managed keys are not offered for disks.

Encryption is set at creation and cannot be changed afterwards, and it is free of cost. It is not available in every region — where it is unavailable, the switch does not appear on the create form.

What Block Storage Does Not Provide​

A disk is deliberately narrower than an Object Storage bucket. There is no S3 endpoint, access key, data-object browser, upload dialog, lifecycle rule, versioning, object lock, presigned URL, or replication. Data goes in and out through the workload the disk is mounted on.

If you need any of those, use Object Storage.

Best Practices for Block Storage

Size it honestly

You are billed on the allocated size, not what you use, and a disk can never be shrunk. Start close to what you need and grow it later.

Use it for what it is good at

Scratch space, caches, and write-heavy steps. Keep large reference data in Object Storage and read it from there.

Mind the single mount

A disk attaches to one workload at a time. For data several workloads read concurrently, use Object Storage, SFS, or PFS.

Back up before deleting

Deleting a disk destroys its data permanently. Copy anything you need into Object Storage first.

Last updated on October 19, 2026.