← Back to Wiki
Mining / HiveOS

Update Nvidia Drivers in HiveOS

The update itself is one command. The parts worth getting right are running it somewhere you can watch it, and checking what it did to your hashrate afterward. A newer driver is not automatically a better one on a mining rig.

Share on X

Put the rig in maintenance mode

Log into your farm and open the rig you are updating.

Go to the settings tab.

Scroll to maintenance mode and choose the option without loading drivers.

The worker updates and needs a reboot to apply it.

Trigger the reboot yourself rather than waiting. It is quicker.

Watch it drop offline and come back.

When it returns it shows as online and in maintenance mode, with the cards offline and no hashrate.

That is what you want. The driver cannot be replaced while it is loaded.

Open a Hive Shell

Select the rig.

Open remote access and start Hive Shell.

Click the link it gives you to open the shell in your browser.

If nothing opens, turn off your browser's shield or ad blocker for the page. Brave blocks it by default.

Turn Hive Shell off again when you are done. It is remote access into the rig and it does not close itself. Leaving it running on every rig you have ever updated is how a convenience feature turns into an exposure.

Run the update

Run it with command nvidia-driver-update

That takes the latest available driver.

You can name a specific version instead when you need to pin or roll back to one.

Let it work. It reboots the rig when it finishes.

Run this from Hive Shell, not from the rig's own console. The console gives you no progress output at all, so a long update is indistinguishable from a hung one. In the web shell you can watch it move.

Bring it back

Once the rig is back online, open it again.

Go to the settings tab.

Scroll down and disable maintenance mode.

Reboot it once more from the farm.

Confirm the new version from the farm view by expanding the rig entry.

Then check what it cost you

Do not assume the update was free.

One driver update measured here dropped 3070s and 3060 Tis by 8 to 9 MH/s each.

The same driver left 3080s and 3090s alone.

Same farm, same overclocks, same day.

So update one rig, compare it against a rig you have not touched, and only roll it out if the numbers hold up.

Being able to pin a specific version is what makes that safe to test.