unlock-gpu.com
Menu
GitHub

Is gpu unlock safe? What it touches, and what it can break

Is gpu-unlock safe? Start with what it does: nothing is flashed, and the register change does not survive a cold boot. A failed probe is a different problem: the card or PCIe segment can remain unavailable, and some failures have no documented host-side fix.

Before you install gpu unlock

  • gpu-unlock is a third-party open-source project and is not affiliated with NVIDIA. This site is not the project’s official website.
  • The only official source is the GitHub repository. The project’s FAQ states there is no official website, no token, and no official installer.
  • Modifying the boot chain and firmware interaction can leave a machine unable to boot or a GPU unusable. Back up first, make sure you can physically cut power, and try it on a non-production machine.
  • Warranty status is unknown: there is no public evidence that it is voided, and none that it is not.

What gpu unlock changes, and what it leaves alone

Two things move — a kernel module loaded from the initramfs, and the drop-in that puts it there. Two things do not: the card’s firmware and the driver already installed on the machine.

  • Changes state

    A kernel module, not a replacement driver

    gpu_unlock loads ahead of nvidia, flips the rate selectors, then resets the card so the stock driver binds to it.

  • Changes state

    One drop-in in the initramfs

    The file your distribution’s package installs is the only thing that makes the module load before the driver.

  • Not touched

    No VBIOS flash

    The card’s firmware is never rewritten, so undoing it involves no flashing step at all.

  • Not touched

    A state a cold boot clears

    The write is replayed on each boot; power-cycling the machine is what resets the card.

[official README] for what the module touches; the uninstall path is the drop-in removal below.

Known gpu unlock failure modes, graded

Five modes are documented in the project’s own reports. Three have an action you can take; two have no documented card-side workaround. Before any of it applies, the card has to be in the gpu-unlock compatibility table— outside it, the module is skipped and none of these modes is reachable.

  • Highthe card can stay unresponsive until the machine loses power
  • Mediumno card-side workaround is documented; wait for upstream
  • Lowa readout artefact or a platform behaviour, not a failed unlock

  • The card stops answering until power is cut

    High
    Trigger
    A failed probe times out (-110) and the device stays unresponsive across reboots.
    Recovery
    Shut down, unplug, start again. A warm reboot does not clear it.
    Self-serve
    Yes, from this machine

    [issue #23]

  • A PCIe switch drops its whole segment

    High
    Trigger
    Downstream port containment fires and the bus segment, not just the GPU, goes away (Xid 79).
    Recovery
    No card-side workaround is documented. Wait for an upstream fix.
    Self-serve
    No — wait for upstream

    [issue #260]

  • An eGPU enclosure times out on the probe

    Medium
    Trigger
    External enclosures hit the probe budget; the report pins it at 1000 µs.
    Recovery
    No documented workaround from the host side. Treat eGPU as unsupported for now.
    Self-serve
    No — wait for upstream

    [issue #38]

  • nvidia-smi shows a frozen SM clock

    Low
    Trigger
    After a successful unlock the clock can sit around 210 MHz while benchmarks run at full speed.
    Recovery
    Ignore it. It is a telemetry artefact, not a failed unlock — check the load, not the clock.
    Self-serve
    Yes, from this machine

    [issue #258]

  • A laptop silently loses the unlock

    Low
    Trigger
    Many laptops power the discrete GPU down when idle, which resets the state the module set.
    Recovery
    Keep the dGPU powered, or accept that results on laptops are mixed.
    Self-serve
    Yes, from this machine

    [official README]

What gpu unlock does to your warranty

Unknown. There is no public evidence that NVIDIA voids a warranty for this, and none that it does not. Treat it as a risk you take, not as a settled question. Anyone who sells a yes or a no on “is gpu-unlock safe” is going past the public evidence.

Spotting an imitation gpu unlock site

Sites that offer a download, an installer, or a “certified” build are not the project. We do not link to them — search the repository name and compare the commit yourself.

The check is cheap on any gpu unlock claim: open VERSION, read the commit that last touched the file a claim comes from, and compare it with whatever number the other page shows. Anything you cannot match to a commit is unverified.

Getting gpu unlock back to stock

  1. Remove the module from the initramfs configuration

    Delete the drop-in your generator reads, or remove the package that owns it.

  2. Rebuild the initramfs and confirm the module is gone

    lsinitramfs on the image you are about to boot, not on the kernel you happen to be running.

  3. Cold boot and check the card reports what it shipped as

    The card itself was never modified, so no firmware step is involved.

Step-by-step per distribution: the gpu-unlock install page. If the card already timed out, read the graded list above first — that case is not a rollback.

Checked VERSION 2.0.0commit 1fc6d62