Skip to content

TIR Block Storage

View as Markdown

A Block Storage disk is a disk of the size you choose that you mount on a TIR Instance or Virtual Machine, which reads and writes it like a local disk. Use it for work that writes into its folder: downloading and unpacking data, transforming it, scratch space, a cache in front of a larger Object Storage dataset, and checkpoints written during training. The portal labels it Block Storage (Disk Dataset) in the side menu and a Dataset on its screens.

A disk appears in the instance at /datasets/<name>, where <name> is the disk’s Dataset Name, and data goes in and out through the instance that mounts it. You can make it bigger whenever your data grows.

Storage Choose it when
Block Storage One instance reads and writes the data, like a local disk
TIR Object Storage Many workloads read the same data, or you want S3 tools, versioning, or presigned URLs
Shared File System Several instances read and write the same files together
Parallel File System Many workers read and write at high throughput, such as large training runs

To work with disks from code, see the Dataset section of the TIR API reference.

In TIR, open your project. In the side menu, under Storage, select Block Storage (Disk Dataset). You see the disks of the current project and region.

Start with Get started with TIR Block Storage, which creates a disk, mounts it on an instance, and puts a file on it. When you see a message, Troubleshooting TIR Block Storage has the fixes. The TIR Block Storage FAQ has short answers on size, billing, and which products take a disk.