Skip to main content

Common GPU Dedicated Server actions

These actions are on the server's Management tab, which appears only while the server is Active: power, IPMI access, and reinstall.

Dash can't see what's running on the server. Before a power action, stop or checkpoint running jobs. Before a reinstall, copy models, checkpoints, and datasets off the server.

Reboot, power off, or power on

  1. On the Power card, check the state. Select refresh if it looks out of date.
  2. Select Reboot or Power off (while Running), or Power on (while Powered off).
  3. Confirm in the dialog. For Reboot or Power off, running jobs stop immediately.

Result: Dash confirms the request and checks the power state until it changes. A GPU server can take a few minutes to boot; if it doesn't come back, use IPMI.

Request IPMI access

You'll need the public IPv4 address of the computer you'll connect from.

  1. On the Console card, select Request IPMI access.
  2. Under Your public IP address, enter that single IPv4 address. Ranges aren't accepted.
  3. Select Request access.

Result: IPMI access granted shows the IPMI address, a temporary Username and Password, Allowed from, and Expires in (2 hours). Copy the credentials; they're shown once. Select Clear IPMI access to close it early. If Dash says Some IPMI access may still be open, contact Customer Care.

Request an operating system reinstall

A reinstall erases everything on the server, including drivers, frameworks, models, checkpoints, and datasets.

  1. On Reinstall operating system, choose an image.
  2. Check I understand this erases all data and snapshots on this server once carried out.
  3. Select Request reinstall and confirm.

Result: Dash shows Reinstall request raised with support. and a ticket code. Note the code, because the link disappears when you leave the page. Nothing changes on the server until our team carries out the request, which you follow in Support. Each request opens a new ticket, so submit it once. Afterward, install your GPU drivers again.