GPU Shows BIOS but Black Screens in Windows: Driver or Hardware Fault? Safe Mode, DDU & Cross-Testing Guide
Key Takeaways
A GPU that displays the BIOS but black-screens in Windows is usually hitting a path the BIOS never exercises: the full WDDM driver initialises every GPU engine, switches resolution and power states. Work from software to hardware — Safe Mode, DDU clean reinstall, then cross-testing — before concluding the card is faulty.
Why BIOS Working Proves Less Than You Think
During boot and in the BIOS setup screen, the graphics card outputs through its firmware display module, not through a real driver:
- Legacy VBIOS (Video BIOS): the traditional firmware display method used with older BIOS and CSM/Legacy modes.
- GOP (Graphics Output Protocol): the display driver built into the card's firmware for UEFI mode, replacing legacy VBIOS so that UEFI firmware can display without CSM.
In this phase the card does the bare minimum — it initialises a basic framebuffer and outputs a low-resolution image. It never touches the 3D engine, never runs memory at full bandwidth, and never performs power-state transitions.
When Windows loads, the full WDDM (Windows Display Driver Model) driver takes over and does several things the BIOS phase never does:
- Fully initialises every GPU engine — 3D, video decode, display engine, power-management unit.
- Switches to the desktop's native resolution and refresh rate — for example, jumping from the BIOS's 1024x768 to 4K 144Hz.
- Reads the monitor's EDID and negotiates the output mode.
- Enables PCIe power saving (Link State Power Management) and dynamic GPU clocks.
A failure at any of these steps means the picture disappears right around the Windows logo or login screen — while the BIOS phase never triggered those code paths at all. So "the BIOS displays" has essentially zero predictive power for these faults.
TDR: Windows' GPU Watchdog Is a Symptom, Not a Diagnosis
TDR (Timeout Detection and Recovery) is Windows' mechanism for an unresponsive GPU: if the GPU does not respond within a period (about 2 seconds by default, adjustable via the `TdrDelay` registry value), the system attempts to reset the display driver (`nvlddmkm.sys` on NVIDIA, `atikmpag.sys` on AMD).
- Reset succeeds → the screen flashes and comes back (the familiar "display driver stopped responding and has recovered").
- Reset fails → a persistent black screen, or a `VIDEO_TDR_FAILURE` blue screen.
The critical insight: TDR is a symptom, not a cause. A driver bug, a power interruption, overheating, an unstable overclock, or a genuine hardware fault can all trigger TDR. Seeing a TDR event and concluding "the card is dead" is a mistake. In this guide TDR entries serve as one diagnostic clue among several, not as a verdict.
The Most Common Human Trigger: Windows Update Driver Conflicts
Windows Update automatically pushes WHQL-certified display drivers (the DCH variant). If the user previously installed a different version manually, leftover files from the two versions can mix and conflict.
The standard procedure, agreed across multiple repair sources:
Disconnect from the network → boot into Safe Mode → clean removal with DDU (Display Driver Uninstaller) → reinstall the official driver.
The network disconnection exists for a specific reason: it stops Windows Update from automatically installing a generic driver mid-process and contaminating the clean environment you just created.
Old Cards, UEFI GOP and Secure Boot: An Important Boundary
Graphics cards from roughly 2016–2017 and earlier mostly carry only legacy VBIOS with no GOP. On newer motherboards running pure UEFI mode — especially with Secure Boot enabled, which requires a signed GOP — these old cards typically cannot show even the BIOS screen. That is the opposite of this article's symptom, but it is an important boundary to state plainly.
Newer cards (generally NVIDIA GTX 10 series and later, AMD RX 5000 series and later) ship with GOP built in and are compatible with pure UEFI and Secure Boot. If you have an old card on a new board and see nothing at all, do not waste time on the driver flow below — that is a firmware-compatibility problem, not a driver problem.
Confirm You Have This Exact Symptom
All four of the following must apply:
- Boot: the motherboard logo and BIOS setup screens display normally.
- The picture disappears during Windows loading — after the logo spinner, around the login screen.
- The PC stays powered on (fans spinning, drive activity light blinking); sometimes the login sound can even be heard.
- Safe Mode usually still displays, because Safe Mode loads only Windows' basic display driver and never loads the full GPU driver.
If there is no BIOS display at all → this is not the symptom covered here. If Safe Mode is black too → suspicion of hardware rises, but the cable, monitor, output port and power supply have not yet been ruled out.
The Diagnostic Flow: Step by Step
Principle: work from software to hardware, from cheap to expensive, and change only one variable at a time.
Step 0 — Confirm Windows Actually Boots
Listen for the login sound or drive activity; press `Caps Lock` and watch whether the keyboard LED responds. A responsive system means Windows is running and only the display output is broken → continue. No response suggests a problem beyond the display, which is outside this guide's scope.
Step 1 — Boot into Safe Mode (via WinRE)
Force the PC off three times during Windows loading to trigger WinRE, then go to Troubleshoot → Advanced options → Startup Settings → Safe Mode.
- Safe Mode has a picture → the fault is almost certainly in the driver or software layer (the full driver never loads, so nothing breaks). Go to Step 2.
- Safe Mode is black too → skip to Step 4 (the hardware/cable branch).
Step 2 — DDU Clean Removal and Reinstall
- In Safe Mode (with networking), download DDU and the card's official driver first.
- Disconnect the network cable / Wi-Fi — this stops Windows Update from auto-installing a driver mid-process.
- Run DDU in Safe Mode and choose "Clean and restart".
- In normal mode, install the official driver: on NVIDIA choose "Custom installation → Perform a clean installation"; on AMD use the Factory Reset option.
- If the latest driver still black-screens, try one or two older versions. Rolling back to an older stable driver is a legitimate diagnostic step, not "being outdated".
How to read the results:
- Normal after DDU → corrupted or conflicting driver; fixed. Consider pausing automatic driver updates from Windows for a while and observing.
- Only a specific version fails → that driver version has a bug; pin a stable version.
- Every version black-screens → go to Step 3.
Step 3 — Rule Out the Software Layer Beyond the Driver
- Uninstall the most recent Windows Update (WinRE → Uninstall updates).
- Turn off Fast Startup (Control Panel → Power Options). Hybrid shutdown lets driver state linger across boots.
- In the BIOS, set PCIe Link State Power Management to off as a test; check whether the motherboard has a BIOS update available (check — do not flash blindly).
- Test with a single monitor at 60Hz, and try a different cable and a different output port (e.g. DisplayPort instead of HDMI).
- Check Windows Event Viewer / Reliability Monitor for `LiveKernelEvent 141`, `VIDEO_TDR_FAILURE`, or `nvlddmkm` / `atikmpag` entries. These help distinguish a driver-reset failure from a straight power drop.
If none of this works → the hardware branch (Step 4).
Step 4 — Hardware Cross-Testing
- Different slot: move the card to another PCIe x16 slot (rules out slot contact or slot power issues).
- Different cable and monitor: rules out the cable and the monitor — EDID or port faults can convincingly mimic "black screen in Windows".
- Different PC: put the card in another known-good machine:
- The same "BIOS displays, Windows black-screens" symptom follows the card → a card hardware problem (but not yet which part — power connector, VRM, memory — that needs detailed bench inspection).
- The card works fine there → the original machine's problem (motherboard, PSU, or system); do not blame the card.
- Different PSU: if power delivery is suspected, cross-test with a borrowed PSU. Note: this is the "PSU wattage or ageing" branch — distinct from the connector-contact branch covered in the companion article on power connectors; do not mix the two.
Step 5 — Firmware-Level Distinction (Judgement Only, No Flashing)
- On cards with a Dual BIOS switch: power off, flip to the other BIOS, and boot. If the other side boots into Windows normally → the VBIOS/GOP firmware is corrupted, not the hardware (the recovery procedure itself is a separate topic and is not covered here).
- Without Dual BIOS, a suspected VBIOS problem is very hard to distinguish from hardware failure by symptoms alone — it requires a repair shop to read the firmware with a programmer. Do not attempt random reflashing.
The Quick Check People Forget: Win+Ctrl+Shift+B and the EDID Branch
Sometimes the machine is fine and the monitor simply received a mode it cannot handle — a too-high refresh rate or a bad EDID. Pressing Win+Ctrl+Shift+B restarts the graphics driver immediately without rebooting. For diagnosis, start from a baseline of a single monitor at native resolution and 60Hz, then add things back one at a time. Cables, adapters, and DisplayPort-version / high-bandwidth (high refresh + DSC) combinations are suspects in this branch too.
Edge Cases
- Old card without GOP on a new motherboard: the typical presentation is *no BIOS display at all* — a different symptom. Enabling CSM is a workaround, but it conflicts with Secure Boot and Windows 11 requirements; the long-term fix is replacing the card. Do not write this off as a "driver issue".
- Dual-BIOS card with one side normal and one side black: a firmware problem, not dead hardware. Do not scrap the card.
- Partial VBIOS corruption: if the damaged area happens to be the GOP region, the BIOS can display under CSM mode while UEFI/Windows black-screens. Rare — list it as an edge case, not the mainstream.
- Mined or second-hand cards with third-party VBIOS: a black screen the moment the driver loads is a common presentation. Asking about the card's history is part of the diagnosis.
- Laptops and all-in-ones: the discrete GPU is soldered; slot-swapping and cross-PC tests are not possible. Rely on an external monitor plus the Safe Mode and driver branches, and go straight to a repair shop for the hardware branch.
Common Misconceptions
| Misconception | The reality |
|---|---|
| "BIOS displays, so the card is fine" | The BIOS only uses a basic framebuffer; it never stresses the 3D engine, memory bandwidth, or power-state transitions |
| "Reinstall the driver" means uninstall and reinstall | Without DDU, leftover files remain; only a clean removal is a valid test |
| "The latest driver is always best" | New drivers can have bugs; rolling back to a stable version is a normal diagnostic step |
| "TDR means the card is dead" | TDR is a symptom; causes include driver bugs, power, heat, and overclocking |
| "Black in Safe Mode means the card is dead" | The cable, monitor, output port, and PSU have not yet been ruled out |
| "Windows Update drivers are the most stable" | Auto-pushed versions may not suit your card — and they reinstall automatically after DDU, contaminating the test |
FAQ
Q: The screen goes black right when Windows starts loading. Is my graphics card dead?
A: Not necessarily. That exact timing — fine during BIOS, dead during Windows load — is precisely when the full driver takes over from the firmware display. Most faults with this timing are resolved in the software steps above (Safe Mode, DDU, driver rollback, recent-update removal). Hardware is only likely when every software step fails *and* cross-testing points back at the card.
Q: Why does Safe Mode still show a picture?
A: Safe Mode loads only Windows' basic display driver, which never touches the 3D engine, power management, or high-bandwidth output modes. Faults hiding in the full driver path — driver corruption, a buggy driver version, aggressive power-state transitions — simply never get triggered.
Q: I ran DDU and the latest driver still black-screens. What now?
A: Try one or two older driver versions before anything else — the latest release may carry a bug affecting your card. If every version fails, work through Step 3 (uninstall the latest Windows Update, disable Fast Startup, test 60Hz with a different cable and port, check the event log), and only then move to hardware cross-testing.
Q: Can a faulty cable really mimic a driver problem?
A: Yes. A bad cable, a faulty port, or a botched EDID negotiation can produce "no signal" exactly when the driver switches resolution and refresh rate — while the PC itself keeps running normally. Always test with a different cable and a different output port before blaming the card.
Q: My old card has no GOP and my new board runs pure UEFI. Will this guide help?
A: That combination usually shows no BIOS display at all, which is the opposite of this guide's symptom. The workaround is enabling CSM, though that conflicts with Secure Boot and Windows 11 requirements; the long-term answer is a card that ships with GOP.
When to Bring It In for Repair
If the diagnostic flow ends at the card itself — or if VBIOS/GOP firmware corruption is suspected but the card has no Dual BIOS switch — the next step is bench-level inspection: slot and connector testing, power-rail measurement, and firmware reading. Services are charged across a published three-tier range of HK$400–2,800, with the final quote confirmed after inspection. The shop is at Room 1209, Dali Plaza, Shek Mun; phone 2186 8010; WhatsApp 6558 6806. There is no on-site service. If you bring the card in, bring the power cable you used with it — the supply chain is part of the evidence.