GPU Dedicated Server troubleshooting
Find the symptom below. Each case lists what to check in Dash, what to do, and when to contact Customer Care. Drivers, CUDA, frameworks, and your jobs run inside your operating system and are yours to fix.
To contact Customer Care, open a ticket, choose the server under Related service or domain, and include the service ID, what you tried, and the exact message Dash showed.
The server is still Pending or In progress
- Check: the status on the servers list, and whether the order's invoice is Paid.
- Do: pay any remaining balance on a Partially paid invoice. Pending and In progress mean our team is setting up the machine. If the order showed an estimated delivery time, use it as a guide. Don't place a second order.
- Customer Care: if the invoice is Paid and you need an update on setup.
The server doesn't come back after a driver or kernel change
- Check: the Power card. Running with no connection usually means the boot stopped.
- Do: request IPMI access and fix the boot from the server's screen. If the operating system can't be recovered, request a reinstall, which erases everything.
- Customer Care: if IPMI shows a hardware error or no screen.
A GPU is missing or reports a fault
- Check: that the GPUs your GPU tools report match the GPU on the server card you ordered, with drivers installed.
- Do: install or update the GPU drivers, then reboot.
- Customer Care: if a GPU is still missing or reports a hardware fault. Include the GPU model you ordered and your GPU tool's output.
A job or framework fails
- Check: the server is Active and Running in Dash.
- Do: if it is, the problem is inside the operating system and outside PerLod's support.
- Customer Care: only if the server isn't Active, doesn't respond to power actions, or shows a hardware fault.
IPMI access is refused or won't connect
- Check: the message in Request IPMI access. It accepts only a single IPv4 address. Access works only from the Allowed from address, for 2 hours, and the password is shown once.
- Do: request access again with the right address. The messages and fixes are the same as on Dedicated Server troubleshooting.
- Customer Care: if the request fails with Something went wrong on our side, access won't connect from the allowed address, or Dash says Some IPMI access may still be open.
Power buttons are greyed out or an action fails
- Check: the Power card badge. Starting up, Shutting down, or State unknown holds all controls; a GPU server can take several minutes to boot. Too many attempts in a row. Wait a moment, then try again. means wait before retrying.
- Do: wait, select the refresh button, and try once more.
- Customer Care: if the state stays unknown, or the same action fails twice.
Reinstall is missing or hasn't happened
- Check: the Reinstall operating system card appears only on an Active server where reinstall is available. No images are offered for this plan. means there's nothing to choose.
- Do: a submitted request is carried out by our team. Follow its ticket in Support, and don't submit it again.
- Customer Care: if the card is missing, or the image you need isn't listed.
The server is suspended
- Check: the header shows Suspended and Scheduled for termination on a date.
- Do: select View invoice on the Overview tab and pay it before that date. The server returns to Active straight away; check your jobs once it's back. See GPU Dedicated Server billing and lifecycle.
- Customer Care: if there's no View invoice link, or the server stays suspended after payment.