GPU Code 43 in Device Manager Doesn't Mean a Dead Card: Driver, System & Hardware Diagnosis

2026-10-20 ｜ GPU Repair ｜ About 12 min read ｜ HK GPU Fix Editorial

Key Takeaways

Windows Code 43 means the system stopped the GPU because the driver reported a problem — a driver-level flag, not a death certificate. Most cases trace to driver corruption, Windows Update driver swaps, or system issues like PCIe seating and UEFI settings. Diagnose in three layers — driver, system, hardware — and reach a verdict on evidence, not panic.

What Code 43 Actually Means

Open Device Manager, expand the display adapters, double-click your graphics card, and on the General tab you may read: *"Windows has stopped this device because it has reported problems. (Code 43)"*. A yellow exclamation mark sits on the card's icon. To many owners, this reads like an obituary. It is not.

The mechanism is worth understanding precisely, because the wording already tells you what is happening. The device itself — via its driver — reported a problem to Windows, and Windows, acting to protect system stability, proactively disabled the device. Code 43 is therefore a driver-level flag: a reporting channel between the driver stack and the operating system. It is not a hardware death certificate.

Two corollaries follow from this:

• Code 43 is not GPU-specific. Microsoft's documentation defines it as a generic device status. It can appear on USB devices, network adapters, and any other hardware Windows manages. It shows up on graphics cards more often simply because the GPU driver stack is unusually complex — more layers means more ways for the handshake to fail.
• Code 43 is the starting point of a diagnosis, not the conclusion. It tells you *that* the driver-to-OS conversation broke down; it does not tell you *why*. The breakdown could be corrupt driver files, a wrong driver version, a badly seated card, a UEFI misconfiguration, a mismatched vBIOS, or genuine hardware failure. Treating all of these the same way is how owners end up replacing a perfectly good card.

The repair logic used by technical communities (petri.com, yellowgnu.com, vividrepairs.co.uk among them) is consistent: work from soft to hard, clearing software causes completely before ever concluding hardware. This article follows that order.

The Three-Layer Diagnostic Framework

Every Code 43 case belongs to one of three layers. Clear them in order, and never skip a layer because you are impatient.

Each layer below ends with its "cleared" signal. If you cannot honestly claim a layer is cleared, you cannot claim the next layer's conclusion.

Layer 1: The Driver Layer — Clear This First

What a driver-layer problem looks like

Three common failure modes live here:

Corrupted installation. If a driver install was interrupted — a power cut, a system crash, or an antivirus blocking part of the process — the result is a hybrid: some files new, some files old. The driver loads, fails its own checks, reports a problem, and Windows disables the card. Code 43.

Incompatible version. Not every driver release is stable on every card-plus-OS combination. A documented case on Microsoft Learn's Q&A involved an RTX 3060 Ti owner whose problems began with the newest release and disappeared after rolling back to an older, stable driver. The lesson: when Code 43 appears, do not blindly chase the newest release. Try one or two previous stable versions instead.

Leftover conflicts. Switching between NVIDIA and AMD (in either direction) leaves residues of the old vendor's driver. The new driver loads into an environment polluted by the old one's registry entries and files, and the conflict surfaces as Code 43. NVIDIA-to-AMD environments need particularly thorough cleanup.

The silent driver swap by Windows Update — the invisible culprit

This is the single most common hidden cause documented across forums, and it deserves its own heading.

The mechanism. Windows Update actively downloads display drivers that Microsoft has tested — most of the time older or generic versions. You may boot up in the morning to discover that Windows quietly replaced your working NVIDIA or AMD driver overnight: a partial overwrite, or an "incomplete update" that replaced only some files. The card that worked yesterday now carries Code 43.

The characteristic pattern. The card was perfectly fine; after a Windows Update or a restart, Code 43 appears out of nowhere. You installed nothing yourself. The driver version in Device Manager has changed on its own — that is the fingerprint.

How to check. Device Manager → your graphics card → Driver tab → note the driver version and date, then compare it against the version published on NVIDIA's or AMD's official site. If they do not match, and you did not install anything, Windows swapped it.

How to stop it happening again (documented by appuals.com, umatechnology.org and vividrepairs.co.uk):

• In Windows Settings → System → About → Advanced system settings → Hardware tab → Device Installation Settings → select No.
• On Pro editions, via gpedit.msc: Computer Configuration → Administrative Templates → Windows Components → Windows Update → "Do not include drivers with Windows Updates" → Enabled.

One operational rule: never run the NVIDIA/AMD installer while Windows Update is connected to the internet — Windows may interfere with the install halfway through.

DDU clean reinstall — the Safe Mode procedure

DDU (Display Driver Uninstaller, from wagnardsoft.com) is the community-standard tool for a true clean reinstall, and the procedure below is consistent across Tom's Hardware forums, Microsoft Learn Q&A, and yellowgnu.com:

1. Download the driver first. Go to NVIDIA's, AMD's, or Intel's official site, download the driver version you intend to install, and place it on the desktop. Do not install through GeForce Experience for this step.
2. Disconnect the network. Unplug the cable or disable Wi-Fi. This prevents Windows Update from interfering mid-process.
3. Enter Safe Mode. Settings → Recovery → Advanced startup, or via msconfig → Boot tab → Safe boot.
4. Run DDU in Safe Mode. Select the correct GPU vendor (NVIDIA / AMD / Intel) and choose "Clean and restart".
5. Install the downloaded driver while still offline. In the NVIDIA installer, choose "Custom" and tick "Perform a clean installation".
6. Restart, reconnect the network, and confirm the exclamation mark is gone in Device Manager.
7. Then lock down Windows Update so it cannot swap the driver again (see the previous section), or the problem will loop.

Two accuracy notes. First: the NVIDIA installer's own "clean installation" option is not equivalent to DDU — it only removes NVIDIA's own files and does not clear registry residues or cross-vendor leftovers. Use DDU whenever a leftover conflict is suspected. Second: the "disconnect from the network" step is consistent community best practice across multiple sources, but it is not a mandate written in DDU's official documentation — treat it as a strong recommendation, not an official requirement.

When the driver layer is cleared

The driver layer is cleared only when you have done a full DDU Safe-Mode reinstall following the steps above, tried two driver versions (including an older stable one), and Code 43 still persists. If that is the case, the problem is not in the driver layer — move to the system layer.

Layer 2: The System Layer — Motherboard, Slot, UEFI

Reseat the card and cross-test a second slot

Power off, unplug the power cable, remove the card, and clear dust from the slot. Clean the gold fingers with an eraser or a clean finger wipe — this is standard guidance across multiple sources. Reinstall the card firmly, and make sure the auxiliary power connectors (6-pin / 8-pin) are fully seated: several Tom's Hardware threads stress that a loose power plug alone can produce Code 43.

If the motherboard has more than one PCIe x16 slot, move the card to the second slot and test. This separates "the card is at fault" from "the slot is at fault". If the second slot works, the problem is the original slot or the motherboard's traces; if both slots show Code 43, the problem is in the card or the power supply.

An honesty note: if you have no second slot and no second machine, this cross-test cannot be done. State the limitation plainly rather than pretending you did it — skipping the test is not the same as passing it.

The PCIe lane clue in GPU-Z

GPU-Z's main page has a "Bus Interface" field showing something like "PCIe x16 4.0 @ x16 4.0". The value before the "@" is what the card supports; the value after is what it is actually negotiating right now.

If an x16 card is negotiating only x8, x4, or even x1, the PCIe link training is failing — which points at poor slot contact, oxidized fingers, physical slot damage, or the motherboard BIOS not supporting this card properly. This is important evidence pointing at the system layer rather than a dead GPU core.

Two caveats, and they matter:

• Many cards downshift lanes and speeds at idle to save power (dropping to x8 at PCIe 1.1 speeds when idle is normal). Click the "?" button in GPU-Z to run the render test and read the under-load negotiated value before drawing any conclusion.
• This inference is reasoned from the PCIe specification and GPU-Z documentation, with indirect forum support — present it as a clue, never as a verdict.

UEFI settings: Above 4G Decoding, CSM, Resizable BAR

These three switches are worth checking because they govern how the card initializes at the firmware level:

• Above 4G Decoding enables 64-bit MMIO addressing for large-VRAM cards. A winraid forum case on an older platform (X79) showed misconfigured ReBAR / Above 4G settings producing a black screen plus Code 43 on driver load. Modern large-VRAM cards generally need this enabled in UEFI mode.
• CSM (Compatibility Support Module). Enabling CSM means booting in Legacy BIOS mode. Resizable BAR requires UEFI boot, CSM disabled, and a GPT system disk. In that same winraid discussion, the system had to boot UEFI (CSM off) for correct 64-bit PCI addressing — some cards fail to initialize under CSM, and the failure presents as Code 43.
• Resizable BAR itself is not the cause of Code 43. The point is the three switches it touches (UEFI boot, Above 4G Decoding on, CSM off). Do not write "enabling ReBAR causes Code 43" — the correct statement is that mismatched UEFI settings can cause the card's initialization to fail, and the symptom is Code 43.

The general guidance: load the BIOS optimized defaults, ensure UEFI boot with CSM disabled (where the card and OS support UEFI), and enable Above 4G Decoding. After changes, reboot twice to give the motherboard a chance to re-train the PCIe link. Note the evidentiary caution: documented cases exist for older platforms, but there is no large body of literature on modern boards — treat these settings as "possibly related and worth checking", not as established causes.

Motherboard BIOS update — last of the system steps, with risk

Multiple sources (vividrepairs.co.uk, Microsoft Learn Q&A, and an Intel official support article) list a motherboard BIOS update as a Code 43 fix step, especially for a new card in an old board where the BIOS does not recognize the card's device ID or initialization sequence. The Intel case is specific: an Intel NUC X15 discrete GPU showing Code 43, with the official fix being a BIOS update via F7 (Intel support article 000089979).

The risk warning is mandatory: a BIOS update interrupted by a power loss can brick the motherboard. This step belongs only after the driver layer is fully cleared and a motherboard-recognition problem is suspected — never as the first thing you try.

When the system layer is cleared

The system layer is cleared when you have reseated the card, tested a second slot (or confirmed there is only one), verified the UEFI settings, and the motherboard BIOS is current — and Code 43 still persists. Now, and only now, does hardware failure become the working suspicion.

Layer 3: The Hardware Layer — Reading the Card Itself

Signals that point to real hardware failure

These are drawn from documented forum cases, not from invented statistics:

• The second-machine cross-test fails too. The card shows Code 43 in a different, known-good machine, while that machine runs a different card fine. This is the strongest single piece of evidence that the fault is in the card.
• GPU-Z reads all zeros, N/A, or Unknown. A documented Tom's Hardware case from January 2026 (an MSI RTX 3090 Ti Suprim X): no display at boot, fans spinning, lights on — but GPU-Z showed all clocks at 0 MHz. The consensus judgment was that the card had died. Note the corollary: fans spinning and lights on do not mean the card is fine — power reaching the fans says nothing about whether the GPU core or VRAM communicate.
• Artifacts appear at boot, before Windows loads. A Microsoft Learn Q&A case combined green lines at boot with Code 43, and no driver reinstall (old versions included) fixed it. Boot-screen artifacts are strong hardware evidence because the driver is not even involved yet at that stage — the failure is at the VRAM or core level.
• GPU-Z shows 0 MB VRAM plus artifacts. Another documented Tom's Hardware case: 0 MB of detected VRAM with purple artifacts that disappeared when dragging windows points at VRAM failure.
• Intermittent Code 43 — fine today, broken tomorrow — may indicate poor contact (system layer) or early-stage component aging. It does not mean the card is definitively dead; do not write it off prematurely.

Counter-signals — evidence it is NOT hardware

Just as important, because they keep you from replacing a healthy card:

• GPU-Z reads the correct model, clocks, and VRAM capacity. If the driver can still interrogate the card accurately, a catastrophic hardware failure is unlikely.
• The problem started right after a Windows Update. The timing alone makes the driver layer the overwhelming suspect.
• A reseat or slot change fixed it. That is a system-layer problem, full stop.
• Disable → Enable in Device Manager fixes it temporarily, and it returns after reboot. This pattern (documented by ittrip.xyz) points at a driver initialization problem rather than hardware.

What this article deliberately does not claim

You will notice there are no "X% of Code 43 cases are driver problems" figures here. None exist in reliable public data; one repair site's claim of "quick reinstall works about half the time" or "DDU fixes about 80%" is that shop's own assertion and cannot be quoted as fact. The honest formulation is qualitative: driver-layer causes are common and usually first. Similarly, this article states no specific voltage or resistance values — board-level measurements are not consumer-article territory and no source supports them.

The Second-Hand Card Special Case: vBIOS Mismatch

Second-hand cards — especially ex-mining cards — deserve their own chapter, because their failure mode sits between software and hardware.

The mechanism

The vBIOS is the firmware on the card, defining its device ID, clocks, power limits, and memory timings. Mining cards were commonly flashed with BIOSes tuned for modified memory timings; resold without being restored to the factory version — or flashed with the wrong vendor's or wrong model's version entirely. A mismatched vBIOS means the motherboard-to-GPU handshake fails at boot, the driver cannot load or refuses to, and the symptom is Code 43. Documented Tom's Hardware cases include a card flashed with an Alienware vBIOS on an ASUS card, and pre-flashed second-hand cards that only recovered with the verified same-vendor BIOS.

The AMD mining special case is instructive: a modified vBIOS fails the BIOS signature check, and AMD's official driver refuses to load — producing Code 43. One MSI forum user worked around it with the third-party atikmdag-patcher, which bypasses the signature check. This article mentions that case only as evidence that the card's BIOS had been modified — not as instructions. The patcher is a signature-bypass tool with legal and warranty grey areas; teaching its use is out of scope.

How to check — without blind flashing

1. In GPU-Z, copy the "BIOS Version" string.
2. Go to TechPowerUp's VGA BIOS database (techpowerup.com/vgabios) and find the BIOS matching your exact brand, exact model, and exact VRAM capacity. Compare the version strings.
3. If the version does not match, vBIOS mismatch is a strong suspect — but only after the driver layer, reseat, and slot tests are done. It ranks behind them.
4. Never blind-flash. Flashing the wrong BIOS, or losing power mid-flash, will genuinely brick the card. Only proceed with a verified match (same vendor, same model, same VRAM), and understand the risk. If you are not confident, do not do it yourself.

One more check direction: if GPU-Z shows a yellow [FAKE] warning before the model name, the card is a fake (a counterfeit device ID was flashed), and no amount of driver work will fix Code 43 — the card needs replacing (documented by appuals.com).

Where vBIOS sits in the suspicion order

Second-hand card + Code 43 + a clean driver reinstall that failed + reseat and slot tests that failed → vBIOS mismatch formally enters the suspect list, as the final software-layer check before professional service.

Myths vs Reality

Before You Send the Card In: A Pre-Service Checklist

A professional diagnosis starts with evidence, not guesses. Whether you attempt further steps yourself or hand the card to a shop, gather the following — it is the same logical flow a repair bench works through:

1. History. When did it start? What happened just before — a driver update, a Windows Update, new hardware? New or second-hand purchase?
2. Software evidence. A Device Manager screenshot, the installed driver version, GPU-Z readings (model, BIOS version, bus interface negotiation, whether clocks read zero), and whether artifacts appear at boot.
3. Driver layer, done properly. DDU Safe-Mode clean reinstall with a known stable driver version — not just an overlay install.
4. Cross-test. Second PCIe slot, then a second machine if one is available. This step is the dividing line.
5. Visual inspection. Any burn marks on the PCB, swollen capacitors, scorched or loose power connectors, oxidized gold fingers.
6. Power supply. Is the PSU wattage and cabling adequate for the card? Describe generally — no invented numbers.
7. vBIOS check for second-hand cards. Match the BIOS version against the TechPowerUp database as described above.
8. Conclusion branches. Driver or system layer fixed it → lock Windows Update and hand the card back. Cross-test confirms a dead card → get a board-level repair quote or replace the card; if it is under warranty, go the RMA route. vBIOS problem → assess whether a safe reflash to the factory version is possible.

Stop and hand it over when you reach hardware-layer suspicion with a second-hand card of uncertain history, when a vBIOS reflash is on the table and you are not confident, or when anything involves opening the card itself. Guessing at the hardware layer is how working cards get destroyed.

FAQ

Q1: Does Code 43 mean my graphics card is dead?

No. Code 43 means Windows stopped the device because the driver reported a problem. It is a driver-level flag, and most cases clear up in the driver layer (corrupt or swapped drivers) or the system layer (seating, slots, UEFI settings). A card is only a hardware suspect after both layers are properly cleared.

Q2: My card worked yesterday and shows Code 43 today. I changed nothing. What happened?

Check whether Windows Update ran overnight — this is the classic signature of a silent driver swap. Open Device Manager → the card's Driver tab and compare the installed driver version against the official NVIDIA/AMD release. If it changed without your involvement, a DDU clean reinstall plus locking Windows Update out of driver installs breaks the loop.

Q3: I reinstalled the driver and Code 43 is still there. Is the card dead?

Not necessarily. An overlay reinstall does not clear leftover conflicts or partial installs. Do a proper DDU clean reinstall in Safe Mode with the network disconnected, and try an older stable driver version too. Only if Code 43 survives that does the investigation move to the system layer.

Q4: The fans spin and the lights are on, so the card must be fine — right?

Wrong. Fans and lighting prove only that power reaches those parts. Documented dead cards spin their fans and light up while GPU-Z shows all clocks at 0 MHz. Judge the card by its GPU-Z readings and behavior, not its decorations.

Q5: Should I try flashing the vBIOS to fix Code 43?

Only as a late, careful step — and never blindly. First confirm the vBIOS is actually mismatched by comparing your GPU-Z BIOS version against the TechPowerUp VGA BIOS database for your exact brand, model, and VRAM capacity. Flashing the wrong BIOS, or losing power mid-flash, will brick the card. If you are not confident, leave it to a professional.

Q6: When should I stop troubleshooting and get professional help?

Stop when the driver layer (DDU reinstall with two driver versions) and the system layer (reseat, slot cross-test, UEFI check, motherboard BIOS) are both cleared and Code 43 remains — or earlier, if the card is second-hand with unknown history, if artifacts appear at boot before Windows loads, or if a vBIOS reflash is the next step. Hardware-layer guesses are how working cards get destroyed.

Need a Professional Diagnosis?

If Code 43 survives the driver and system layers, if the card shows boot-level artifacts, or if a second-hand card needs its vBIOS checked against the factory version, chip-level diagnosis is the next step. WhatsApp 6558 6806, call 2186 8010, or visit the shop at Room 1209, Dali Plaza, Shek Mun. Repairs run on three published tiers (HK$400–2,800), with the final quote confirmed only after inspection.