GPU Keeps Restarting in a Boot Loop? Boot Stage, Driver Load & Power Protection Diagnosis Guide

2026-11-11 | GPU Repair | 24 minutes | HK GPU Fix Editorial Team

GPU Keeps Restarting in a Boot Loop? Boot Stage, Driver Load & Power Protection Diagnosis Guide

Key Takeaways

A PC that powers on, then restarts itself before reaching a stable desktop, is caught in a boot loop, and the stage where the restart happens already narrows the cause. Windows may be hitting a serious error and restarting automatically, or power-supply protection may be cutting power and trying again.

What Counts as a Boot Loop

Pressing the power button starts the machine — fans spin, lights come on, the motherboard logo or the Windows loading screen may appear — but before a stable desktop is reached, the machine restarts itself, then starts again, then restarts again, and keeps going. Three features must all be present before the word "boot loop" fits:

  1. The machine does power on. The boot chain has started at least once. A machine that shows no sign of life at all is a different problem.
  2. The restart happens on its own. Nobody pressed the restart button. The system restarts itself.
  3. It repeats. This is not a one-off restart. Every boot ends in a restart at roughly the same point, or in a restart somewhere along the way.

The single most useful piece of information is where the restart happens. Note the exact stage before changing anything, because the stage itself is a locating tool. A restart at the same point every time and a restart at a different point every time point in opposite directions, and that contrast is the starting point for everything that follows.

Where the restart happensWhich layer it points to
Before or at the motherboard logo / POST screenFirmware, power supply, or hardware self-test. The graphics driver has not loaded yet, so the driver is not the first suspect
At the Windows loading screen, while the spinner is turningBoot driver loading or boot file loading
At the sign-in screen or lock screenThe driver load point and desktop composition layer
Shortly after reaching the desktopDriver, service, or startup program layer — things that only load after sign-in
At the same point every timeA deterministic restart. The same trigger condition repeats: the same driver, the same set of boot files, or the same setting
At a different point each time, seemingly at randomA non-deterministic restart. Randomness is itself evidence, pointing towards power supply, heat, memory, or connection problems

A deterministic restart — the same point every time — means one trigger condition is firing again and again. A random restart means the trigger is not tied to a particular step in the boot sequence. Write down which of the two it is, with the exact stage, rather than writing "about halfway through". Precision here saves every later step from chasing the wrong layer.

Two Different Restart Paths, Plus a Repair Loop

A boot loop is not one mechanism. Two different paths produce the same outward behaviour, and they need to be separated before any diagnosis can begin.

Path One: Windows Hits a Serious Error and Restarts Automatically

When Windows hits a serious error — a stop error, also called a bug check or blue screen — its default behaviour is to restart automatically. Automatic restart is a setting, not a fault in itself. The fault is whatever keeps triggering the error on every boot: a faulty boot driver, a failed update, or damaged boot files. Each boot runs into the same error, Windows restarts, and the cycle repeats: boot, error, automatic restart, boot again, error again.

The restart itself is not the fault. The restart is Windows responding to a serious error in the way it is set to respond. Finding the fault means finding what triggers the error, and that starts with making the error screen stay visible long enough to read it.

Path Two: Power-Supply Protection Cuts Power and Tries Again

A power supply unit carries several internal protections: over-voltage, under-voltage, over-current, over-power, short-circuit, and over-temperature. When one of them trips, the power supply cuts power to protect itself. In some cases the system then tries to power on again, producing a cycle of "on for a moment, off, on again, off again".

The difference from path one is stark: there is no blue screen, no error code, and no Windows screen at all. The machine cuts power directly instead of going through a Windows restart. After protection has latched, many power supplies need a complete disconnection from the wall — unplug the power cord and wait for residual charge to drain — before they will reset. If pressing the power button immediately after a failure does nothing at all, and the machine only powers on again after the cord has been unplugged for a while, that fits a protection latch waiting to reset.

The Automatic Repair Loop

After two consecutive failed boots, Windows enters Automatic Repair on its own. The screen shows "Preparing Automatic Repair", then "Diagnosing your PC", then reports that it could not repair the PC — "Automatic Repair couldn't repair your PC" or "Your PC did not start correctly" — and offers Restart and Advanced options. Pressing Restart sends the machine back into Automatic Repair, and the loop continues.

Automatic Repair can also be triggered by two crashes or unexpected shutdowns within two minutes of booting. It is distinct from path one: in path one, Windows is still trying to boot normally and hits an error on the way; in the Automatic Repair loop, Windows has already given up on a normal boot and moved into repair mode, and the repair itself is failing. The two must not be lumped together under "won't boot".

The First Cut: Stop Automatic Restart and Write Down the Stop Code

This is the most important single step in the whole diagnosis, and its order cannot be rearranged. Until automatic restart is switched off, any statement about which driver or which part is at fault is a guess, because the error screen flashes past too quickly to read.

How to Switch It Off

Enter the Windows Recovery Environment (WinRE), then go to Troubleshoot, Advanced options, Startup Settings, Restart, and choose option 9, "Disable automatic restart on system failure". After that, wait for the machine to hit the error once more. The error screen will stay on the display. Photograph it or write down the complete stop code and the module name listed after "What failed", if one is shown.

Getting Into WinRE

BitLocker Warning — Before Entering WinRE

Before changing startup settings, locate the BitLocker recovery key first, if the machine uses BitLocker or Windows 11 device encryption. Without the key, the repair tools may not be able to read the Windows drive. This warning belongs before any WinRE step, not after it.

How to Use the Stop Code

A stop code says which part of the system reported the error. It does not say which part is dead. The four codes below are the only ones listed. Use only the code that actually appears on your screen — never assume a particular code will appear. Never claim a particular code will appear.

Stop codeNameWhich area reported it
0x116VIDEO_TDR_FAILUREDriver timeout reset failure
0x119VIDEO_SCHEDULER_INTERNAL_ERRORDisplay scheduler internal error
0x10EVIDEO_MEMORY_MANAGEMENT_INTERNALVideo memory management internal error
0x113VIDEO_DXGKRNL_FATAL_ERRORDirectX graphics kernel fatal error

A code is a report from one department, not a conviction of one part. It must be combined with the Safe Mode test and the change timeline before it narrows anything down. Event 4101, the "display driver stopped responding and has recovered" event, applies only when the driver actually recovered. In a boot loop the driver never recovered, so Event 4101 is not evidence for a loop.

Windows 11 24H2 and Later: The Crash Screen Changed

From Windows 11 24H2 onwards, the crash screen is black rather than blue, and the sad face and QR code are gone, with a shorter message. A black screen does not mean the error is not a blue screen. The stop code is still on the screen. Write it down.

The Second Cut: Safe Mode Separates the Driver Layer from the Hardware Layer

In WinRE, go to Troubleshoot, Advanced options, Startup Settings, Restart, and choose option 4, Safe Mode — or option 5, Safe Mode with Networking, if downloading a driver is needed. Safe Mode loads only the most basic drivers and services.

Safe Mode resultWhich layer it points toNext step
Safe Mode is stable, no more loopingDriver, service, or startup program layer. Something that only loads during a normal boot is the suspectFollow the driver-layer branch below
Safe Mode loops as wellHardware, firmware, or boot file layer. The cause survives even the minimal load of Safe ModeFollow the hardware branch and the Automatic Repair branch below
Safe Mode cannot be entered at allThe problem sits earlier: boot files, storage, or firmwareStop, and move towards the Automatic Repair and check-in direction

Safe Mode being stable proves only that the problem follows something that loads during a normal boot and not during Safe Mode. It does not identify which driver, and it is not a repair. Entering Safe Mode must never be described as the problem being fixed.

The Driver-Layer Branch: The Recent Change Timeline Is the Strongest Evidence

When Safe Mode is stable, the investigation follows the timeline. What changed most recently before the problem started is the most important single clue.

The loop started after a graphics driver update. In Safe Mode, roll back or uninstall the graphics driver, then install a version that previously worked normally. A clean reinstall follows the DDU plus clean-install procedure. Version choice is judged by the timeline — "newest" is not the same as "best", and an update that arrived just before the problem started is itself a suspect to be rolled back and compared.

The loop started after a Windows update. In WinRE, go to Advanced options, Uninstall Updates. Try "Uninstall latest quality update" first, for the monthly update. If the problem followed a version upgrade, try "Uninstall latest feature update". After roughly ten days, the files needed to roll back a feature update are usually already gone.

The loop started after hardware was changed or a device was added. Remove the newly added hardware and test again. After switching graphics card brands — NVIDIA to AMD, or the other way — leftover driver residue from the previous brand is a known branch to clear.

Nothing changed, and the loop started suddenly. The timeline itself is the evidence. "Nothing changed" points towards the hardware, power, or heat layer more than the driver layer. Move to the hardware branch below.

System Restore. System Restore is one option on the WinRE menu: a restore point takes the system back to an earlier state, does not delete personal files, but removes drivers and programs installed after the restore point. It is a Windows recovery tool, not a graphics diagnosis step — treat it as one recovery option, not as the main step.

The Hardware and Power Branch: No Blue Screen, Direct Power Cut

When the loop shows no blue screen and no error code at all, and the machine simply cuts power and starts again, the suspicion moves to the hardware layer.

The fans spin for a moment, then stop. That is a typical sign of power-supply protection tripping. Before removing the graphics card, disconnect power completely; never touch hardware while the machine is powered. The comparison is to remove the graphics card and test: if the machine powers on reliably every time with the card removed, suspicion sits on the graphics card power side, such as a short; if it still loops with the card removed, suspicion moves towards the power supply or the motherboard.

Protection latched and not yet reset. Pressing power immediately after a failure gives no response at all, and the machine only powers on after the wall cord has been unplugged for a while. That fits a power-supply protection latch waiting to reset.

Power connectors and cables. After disconnecting power, reseat each power cable firmly and check for scorching, discolouration, or deformation. If any scorch mark is visible, stop immediately and do not power on again.

Memory. Memory testing is one of the steps in loop diagnosis. No single testing tool is the only standard, and do not attempt any powered measurement beyond removing and reseating memory.

Storage. A failing drive can make Automatic Repair repeat the same problem endlessly, and a loose storage cable is also a possibility. Check the drive and its cables for looseness; a storage repair is a separate job.

Heat. Heat can only be discussed if the machine stays on long enough to get hot. If it restarts within a few seconds of powering on, there has not been time for heat to build up, so heat is not the first suspect. Do not judge by a universal temperature figure — different configurations have different requirements.

The Automatic Repair Loop Branch: Read the Log, Do Not Keep Re-Running

When the machine is already in an Automatic Repair loop:

  1. Do not keep re-running Startup Repair. Once Startup Repair reports that it cannot repair the PC, running the same repair again will not change the result. Move to the next step instead.
  2. Read the SrtTrail.txt log. The Startup Repair log file records why the repair failed. Its contents can be handed to inspection staff. Interpreting the technical detail inside the log to reach a verdict is inspection work, not something to do at home.
  3. Uninstall Updates. See the driver-layer branch above: quality update first, feature update second, with the roughly ten-day limit on feature rollback.
  4. System Restore and Safe Mode. See the earlier sections.
  5. A boot configuration message on screen. If the screen shows "Recovery — Your PC/Device needs to be repaired" together with 0xc000000f, 0xc0000225, or 0xc000000e, that text as seen on screen indicates damaged boot configuration data or a system that cannot be found. These strings are words seen on the screen, not a diagnosis by themselves, and rebuilding boot records from the command line is a job for inspection, not for home.
  6. Automatic Repair sitting still for more than 30 minutes with no movement at all. That is not a loop. It is a different problem — a stall — and this diagnosis flow does not apply to it.

Diagnosis, Step by Step

The steps run from cheapest to most disruptive. Each step changes one thing, then tests, then records the result.

Step 0: Record the Stage and the Pattern

Write down the exact stage where the restart happens, and whether it happens at the same point every time or at a different point each time. Also note whether any blue screen flashes past, and whether the machine cuts power directly or goes through a Windows restart.

Step 1: Switch Off Automatic Restart and Read the Stop Code

Follow the WinRE path to Startup Settings, option 9. Let the error happen once more, photograph the screen, and write down the complete stop code and any module name shown after "What failed".

Step 2: Test Safe Mode

Enter Safe Mode from Startup Settings. If it is stable, the driver layer is the branch. If it loops too, the hardware and boot file layers are the branch. If it cannot be entered at all, the problem sits earlier.

Step 3: Follow the Timeline

Match the start of the problem against the most recent change: graphics driver update, Windows update, or new hardware. Roll back or remove that change and test again, one change at a time.

Step 4: Test the Power Branch

If there is no blue screen at all, test the power branch: remove the graphics card and compare, check for a protection latch by unplugging the wall cord, and inspect power cables for scorching or damage.

Step 5: Handle the Automatic Repair Loop

If Automatic Repair is looping, do not re-run Startup Repair. Read the SrtTrail.txt log, try Uninstall Updates, and note any boot configuration message shown on screen.

Step 6: The Check-In Boundary

If Safe Mode loops, if the machine cuts power with no blue screen and the card-removal comparison changes nothing, if a scorch mark is visible or you smell burning, or if the Automatic Repair loop persists after the steps above, stop and bring the record to a check-in.

Boundary Cases

Not every machine that fails to boot is in a boot loop. These look-alikes belong to different diagnoses:

SituationHow to tell it apartWhere it belongs
Stuck on the motherboard logo, not moving, no restartStuck still versus restarting in a loop are opposite actionsA stuck-at-logo diagnosis
Takes several presses to power on, then runs normally once onFailing to power on versus powering on and then restarting itself — opposite directionsA power-on failure diagnosis
Restarts only during gaming, fine when idleThe timing is after entering a game, under loadA load-stage restart diagnosis
The loop only starts after pressing Windows Restart; a cold boot is fineThe trigger is a user-pressed restart, not a restart the system made on its ownA restart-trigger diagnosis
Reaches the desktop and restarts after a while of useLeans towards a load or idle crash, not a boot loopA load or idle crash diagnosis
A blue screen appears once, then the machine runs normallyA single crash does not need the loop flowNote the code and observe
The monitor itself switches off or restarts, while the PC keeps runningThe monitor's own protection circuit restarting, not the PC restartingA monitor restart diagnosis
Automatic Repair screen sits still for more than 30 minutes with no movementA stall, not a loopCheck-in direction; this flow does not apply
Boot takes a long time to spin, but eventually reaches the desktopSlow, not loopingA slow-boot diagnosis

Common Misconceptions

What to Note Down Before a Check-In

Bringing a clear record makes a check-in faster and more accurate. Note the exact stage where the restart happens, and whether it is the same point every time or a different point each time. Note whether a blue screen appears or the machine cuts power directly. Note the complete stop code and any module name shown after "What failed". Note the Safe Mode result, the most recent change before the problem started, and the result of each step tried — driver rollback, update uninstall, card-removal comparison, and power cable inspection. Note any message shown on the Automatic Repair screen, including the SrtTrail.txt log if it was read.

Inspection is charged under the site's published three-tier fee of HK$400–2,800, with the actual figure confirmed after inspection. The shop is at Room 1209, Dali Plaza, Shek Mun. The phone number is 2186 8010 and WhatsApp is 6558 6806. There is no shop in Mong Kok and no on-site service.

Frequently Asked Questions

What exactly is a boot loop?

A machine that powers on — fans spin, lights come on, the logo or loading screen may appear — but restarts itself before reaching a stable desktop, and then repeats that cycle on every boot. Three features must all be present: the machine does power on, the restart happens on its own with nobody pressing anything, and it repeats rather than happening once.

Does a boot loop mean my graphics card is broken?

Not necessarily. A boot loop has three possible routes — driver layer, boot file layer, and power protection layer. Note the stage where the restart happens, switch off automatic restart to read the stop code, and use the Safe Mode comparison before blaming the graphics card.

How does the stage where the restart happens help?

The stage is a locating tool. A restart before or at the motherboard logo points to firmware, power, or hardware self-test, because the driver has not loaded yet. A restart at the Windows loading spinner points to boot driver or boot file loading. A restart shortly after reaching the desktop points to things that only load after sign-in. Note the exact stage before changing anything.

My PC restarts at the same point every time. What does that mean?

It is a deterministic restart: the same trigger condition fires on every boot — the same driver, the same set of boot files, or the same setting. That is useful evidence, because it narrows the search to one repeating trigger. A restart at a different point each time is the opposite: non-deterministic, pointing towards power, heat, memory, or connection problems.

My PC restarts at a different point each time. What does that mean?

Randomness is itself evidence. A restart that is not tied to a particular step in the boot sequence points away from a single faulty driver or boot file and towards the hardware layer — power supply, heat, memory, or connections. Do not treat "no pattern" as "no information".

Why do I need to stop automatic restart before I can read the stop code?

Until automatic restart is switched off, the blue screen flashes past too quickly to read, so any statement about which driver or part is at fault is a guess. Enter WinRE, go to Troubleshoot, Advanced options, Startup Settings, Restart, and choose option 9 to switch it off. Then let the error happen once more and photograph the screen, writing down the complete stop code and any module name shown after "What failed".

How do I get into the Windows Recovery Environment?

If the sign-in screen can be reached, hold Shift while clicking Power, then Restart. If the system cannot be reached at all, interrupt the boot with the power button twice during startup, and Windows will enter WinRE on the third boot. From inside Settings, the path is System, Recovery, Advanced startup, Restart now. Before changing startup settings, locate the BitLocker recovery key first if the machine uses BitLocker or device encryption.

What does the stop code tell me?

It tells which area of the system reported the error — for example, 0x116 is a driver timeout reset failure, 0x119 is a display scheduler internal error, 0x10E is a video memory management internal error, and 0x113 is a DirectX graphics kernel fatal error. The code alone does not say which part is dead. It has to be combined with the Safe Mode test and the change timeline.

Safe Mode is stable. Does that mean the problem is fixed?

No. Safe Mode is a comparison test, not a repair. If Safe Mode is stable, the problem follows something that only loads during a normal boot — a driver, a service, or a startup program. If Safe Mode loops as well, the cause is in the hardware, firmware, or boot file layer. Entering Safe Mode never means the machine is fixed.

The loop started right after a driver update. What do I do?

In Safe Mode, roll back or uninstall the graphics driver, then install a version that previously worked normally. Version choice is judged by the timeline — "newest" is not the same as "best". If the loop started after a Windows update instead, use Uninstall Updates in WinRE, trying the latest quality update first.

There is no blue screen at all — the machine just cuts power and starts again. What is that?

That fits the power-protection path rather than the Windows error path. A power supply's internal protections — over-voltage, under-voltage, over-current, over-power, short-circuit, over-temperature — can cut power and then try again. The comparison is to remove the graphics card and test: if the machine powers on reliably with the card removed, suspicion sits on the graphics card power side; if it still loops, suspicion moves to the power supply or motherboard.

Automatic Repair keeps saying it cannot repair my PC. Should I run it again?

No. Once Startup Repair reports that it cannot repair the PC, running the same repair again will not change the result. Read the SrtTrail.txt log, try Uninstall Updates, and note any boot configuration message on screen, such as "Your PC/Device needs to be repaired" with 0xc000000f, 0xc0000225, or 0xc000000e. Those strings are words seen on screen, not a diagnosis by themselves.

Why should I not reinstall Windows first?

WinRE offers Startup Repair, Uninstall Updates, System Restore, and Safe Mode, none of which touch personal files. The order runs from lowest risk to highest risk, and reinstalling is a last resort, not a first step. Reinstalling before separating the layers may not fix the loop, and it wipes away the clues that locate the fault.

Is pressing the power button several times to start the same thing as a boot loop?

No. Pressing the power button several times before the machine starts is failing to power on; a boot loop is powering on and then restarting before reaching the desktop. Opposite directions, completely different flows.

When should I stop and send it in for a check-in?

When Safe Mode loops as well, when the machine cuts power with no blue screen and the card-removal comparison changes nothing, when a scorch mark is visible on a power cable or you smell burning, or when the Automatic Repair loop persists after the steps above. Bring the exact restart stage, the stop code, the Safe Mode result, the recent change timeline, and the result of each step tried. Inspection is charged under the site's published three-tier fee of HK$400–2,800, with the actual figure confirmed after inspection. The shop is at Room 1209, Dali Plaza, Shek Mun. The phone number is 2186 8010 and WhatsApp is 6558 6806.

Related articles

Related services

If you need Power connector repair, see these service pages: