Interactive Troubleshooter

Published: Sep 3, 2026

RTX 5090 / RX 8900 Driver Crash in Games: Fix GPU Driver Timeout

Time: 15 mins Level: Beginner

By StepDiag Engineering Team

Reviewed by Ibraheem Taofeeq Opeyemi

An exhaustive interactive diagnostic guide to fix driver crashes, TDR timeouts, and black screen system hangs on RTX 5090 and RX 8900 series GPUs. Follow structured troubleshooting paths for power transient drops, PCIe Gen 5 instability, and graphics stack conflicts.

Full Diagnostic Tree & Step-by-Step Overview

What specific error code, crash pattern, or system behavior occurs when your graphics card crashes in games?

Black screen with GPU fans ramping to 100%, PC remains powered on and requires a hard reset.
Game freezes and exits to desktop with 'AMD Software detected a driver timeout' or 'Nvidia driver stopped responding'.
The entire system crashes to a Blue Screen of Death (BSOD) referencing 'nvlddmkm.sys' or 'amdkmdag.sys'.
Game engine terminates abruptly with 'DirectX 12 Device Removed' or 'Unreal Engine DXGI_ERROR_DEVICE_HUNG'.

What physical hardware configuration or power delivery pathway is supplying your graphics card?

GPU is powered via a direct 12V-2x6 / 12VHPWR native cable or multi-8-pin PCIe power adapters.
GPU is installed on a PCIe vertical mount, PCIe extension riser cable, or PCIe Gen 5 slot.
GPU operates under aggressive factory overclocking, high ambient room temperatures, or intensive ray tracing.

Inspect 12V-2x6 Cable Seating & Mitigate Power Supply Transient Spikes

Solution:

Root Cause: High-Power Connector Micro-Gaps or Power Supply Transient Excursions

Next-generation flagship GPUs like the RTX 5090 and RX 8900 exhibit extreme transient power spikes (excursions) up to 600W+. If the 12V-2x6 or 16-pin connector sense pins detect high resistance due to a 0.5mm gap, or if the PSU cannot handle microsecond voltage drops on the 12V rail, the GPU safety microcontroller triggers an emergency hardware shutdown, spinning fans to 100% while dropping video output.

Step-by-Step Fix:

1.Inspect and Fully Seat 12V-2x6 / 16-Pin Power Connector:

Shut down the system and disconnect the AC power cord.

Unplug the 12V-2x6 high-power cable from the GPU. Check terminal pins for thermal discoloration or bent sense pins.

Re-insert the plug firmly until an audible latch click is heard. Verify zero visible gap between the male plug and female GPU receptacle header.

Ensure the cable has at least 35mm of straight clearance before any severe lateral bends.

2.Eliminate Daisy-Chained PCIe Cables:

If using 8-pin to 16-pin power adapters, use separate, dedicated 8-pin PCIe cables directly from the Power Supply Unit (PSU).

Avoid splitters or daisy-chained 'pigtail' cables.

3.Verify ATX 3.1 / PCIe 5.0 PSU Overhead:

Confirm your PSU is rated for at least 850W–1000W+ with native ATX 3.0 / 3.1 compliance to absorb 200% transient power excursions without tripping Over Current Protection (OCP).

Fix PCIe Gen 5 Signal Integrity Failure & Force Gen 4 Link Speed in BIOS

Solution:

Root Cause: PCIe 5.0 Signal Attenuation & Riser Cable Impedance Misalignment

PCIe Gen 5 operates at extreme high-frequency bandwidth (32 GT/s per lane). PCIe extension riser cables, dirty motherboard slot contacts, or minor trace impedance mismatches cause signal reflection errors. When packet transmission errors exceed retry thresholds, the PCIe bus drops link negotiation, resulting in an instant black screen driver failure.

Step-by-Step Fix:

1.Force PCIe Generation Speed in Motherboard UEFI/BIOS:

Reboot your PC and repeatedly press Del or F2 to enter the UEFI BIOS setup.

Navigate to Advanced > PCIe Subsystem Configuration (or Onboard Devices Configuration).

Locate PCIe Slot Link Speed (Primary x16 slot) and change the value from Auto or Gen5 to Gen4 or Gen3.

Save settings (F10) and restart into Windows.

2.Bypass PCIe Riser Extension Cables:

Remove vertical GPU mounting brackets or riser cables temporarily.

Install the graphics card directly into the top primary motherboard PCIe x16 slot to isolate physical bus signal degradation.

3.Clean Gold Finger Bus Contacts:

Remove the GPU and gently clean the PCIe gold fingers using 99% isopropyl alcohol and a lint-free microfiber cloth. Use compressed air to blow debris out of the motherboard slot.

Mitigate GDDR7 / VRAM Thermal Throttling & Stabilize Voltage Peak Transient Curve

Solution:

Root Cause: High VRAM Junction Temperatures or Factory V/F Curve Instability

High-density video memory modules (GDDR7 / GDDR6X) generate substantial thermal density under heavy ray tracing workloads. If memory junction temperatures exceed 105°C, or if factory overclocks request voltages that exceed the silicon's stable frequency curve during rapid load transitions, the GPU processor encounters memory parity errors and crashes.

Step-by-Step Fix:

1.Monitor Thermal Sensors with HWInfo64:

Launch HWiNFO64 (Sensors-Only mode) while running a GPU benchmark.

Check GPU Temperature, GPU Hotspot Temperature, and GPU Memory Junction Temperature.

If Hotspot exceeds 100°C or Memory Junction exceeds 105°C, clean thermal dust filters and increase case intake airflow.

2.Apply Voltage/Frequency Curve Undervolt:

Download and install MSI Afterburner.

Press Ctrl + F to open the Voltage/Frequency Curve Editor.

Flatten the core clock curve at 950mV to 1000mV at a target frequency slightly below maximum factory boost (e.g., 2650 MHz).

Save profile and test for load stability in heavy DirectX 12 games.

3.Enable Frame Rate Limits to Cap Power Spikes:

In NVIDIA Control Panel or AMD Software Adrenalin, set a Global Max Frame Rate matching your monitor's high refresh rate (e.g., 141 FPS for 144Hz G-Sync/FreeSync display) to prevent unnecessary power draw in game menus.

When does the display driver timeout or app crash freeze occur?

Crashes began after updating drivers over an existing installation or upgrading Windows.
Crashes occur during initial game shader compilation, fast traveling, or scene loading transitions.
Crashes occur when running background video streams, Discord streams, or web browsers on secondary monitors.

Execute Display Driver Uninstaller (DDU) Routine in Windows Safe Mode

Solution:

Root Cause: Driver Store Corruption & Corrupted Shader Cache Schemas

Overwriting existing graphics drivers using standard installers leaves residual DLL binaries, conflicting registry entries, and outdated shader cache structures in C:\ProgramData. This leads to pointer mismatch errors during runtime dynamic link library calls, triggering a driver timeout reset.

Step-by-Step Fix:

1.Download Required Utilities:

Download the latest Display Driver Uninstaller (DDU) utility.

Download the official WHQL driver package from NVIDIA Official Driver Downloads or AMD Official Drivers and Support.

2.Boot Windows into Safe Mode:

Hold Shift while clicking Restart in the Windows Start Menu.

Select Troubleshoot > Advanced Options > Startup Settings > Restart.

Press 4 or F4 to enter Safe Mode.

3.Execute DDU Removal Process:

Disconnect your network cable / disable Wi-Fi to prevent Windows Update from automatically downloading generic drivers.

Launch DDU, set device type to GPU, and select your GPU vendor (NVIDIA/AMD).

Check options to remove C:\NVIDIA or C:\AMD extraction folders.

Click Clean and Restart.

4.Install Clean Driver Package:

After Windows boots, run the freshly downloaded official setup package. Choose Custom (Advanced) Installation and check Perform a clean installation if available.

Adjust Windows TDR Delay Registry Values & Purge DirectX Shader Cache

Solution:

Root Cause: Windows Timeout Detection and Recovery (TDR) Preemptive Execution

Windows monitors graphics hardware execution using TDR. If the GPU takes longer than 2 seconds (default) to process a complex rendering instruction or compile shaders during high-resolution asset loading, Windows assumes the GPU hardware is hung, resets the driver scheduler, and terminates the gaming application.

Step-by-Step Fix:

1.Modify TDR Delay Parameters in Windows Registry:

Press Win + R, type regedit, and press Enter.

Navigate to the following key:

HKLM\SYSTEM\CurrentControlSet\Control\GraphicsDrivers

Right-click in the right pane, select New > DWORD (32-bit) Value, and name it TdrDelay.

Double-click TdrDelay, set Base to Decimal, and set Value data to 10 (increases threshold from 2 seconds to 10 seconds).

Create another DWORD (32-bit) Value named TdrDdiDelay, set Base to Decimal, and Value data to 10.

Review official parameter guidelines via Microsoft Official TDR Registry Documentation.

2.Purge Corrupted DirectX Shader Cache:

Press Win + I to open Settings > System > Storage > Temporary Files.

Check DirectX Shader Cache and click Remove Files.

For NVIDIA users, open C:\Users\<Username>\AppData\Local\NVIDIA and delete contents in DXCache and GLCache folders.

Disable Multi-Plane Overlay (MPO) & Hardware-Accelerated GPU Scheduling Conflicts

Solution:

Root Cause: Desktop Window Manager (DWM) Multi-Plane Overlay (MPO) Plane Switching

Multi-Plane Overlay (MPO) allows graphics hardware to draw independent screen planes (such as a video stream over a 3D game) to reduce display latency. However, driver bugs in high-bandwidth GPUs cause display timing sync mismatches between hardware acceleration in background apps (Discord, Chrome, OBS) and the main game rendering pipeline.

Step-by-Step Fix:

1.Disable MPO via Windows Registry Patch:

Press Win + R, type regedit, and navigate to:

HKLM\SOFTWARE\Microsoft\Windows\DWM

Right-click, select New > DWORD (32-bit) Value, and name it OverlayTestMode.

Double-click OverlayTestMode and set its value to 00000005 (Hexadecimal) to disable MPO plane composition.

2.Toggle Hardware-Accelerated GPU Scheduling (HAGS):

Open Settings > System > Display > Graphics > Change default graphics settings.

Toggle Hardware-accelerated GPU scheduling to Off (or test On if currently off) and restart the PC.

3.Disable Hardware Acceleration in Background Apps:

In Discord: Settings > Advanced > Disable Hardware Acceleration.

In Google Chrome / Edge: Settings > System > Disable Use graphics acceleration when available.

What auxiliary error symptoms accompany the system BSOD or crash?

BSOD shows error codes like SYSTEM_SERVICE_EXCEPTION or MEMORY_MANAGEMENT alongside GPU driver files.
BSOD occurs only under combined heavy CPU + GPU loads, or memory training fails during cold boots.

Resolve DDR5 Memory Instability (EXPO/XMP) & Resizable BAR Memory Mapping

Solution:

Root Cause: High-Speed System RAM Instability & Resizable BAR Address Overlap

When Resizable BAR (ReBAR) or Smart Access Memory (SAM) is enabled, the CPU maps the entire GPU VRAM framebuffer into main system memory space over the PCIe bus. If high-frequency DDR5 RAM (e.g., 6000MHz+ XMP/EXPO) experiences minor bit errors during memory access, the GPU driver encounters memory management BSODs (nvlddmkm.sys / amdkmdag.sys).

Step-by-Step Fix:

1.Run System Memory Diagnostic Test:

Download MemTest86 or use built-in Windows Memory Diagnostic (mdsched.exe).

Run a full memory pass to check for RAM parity errors.

2.Adjust DDR5 Memory Profiles & Timings:

Enter Motherboard UEFI BIOS.

Temporarily disable XMP / EXPO and run RAM at JEDEC baseline speeds (e.g., 4800MHz) to check if GPU crashes cease.

If stable, re-enable XMP/EXPO but manually set DRAM VDD/VDDQ Voltage up by 0.02V or lower clock speed by 200MHz (e.g., set 6000MHz profile to 5800MHz).

3.Toggle Resizable BAR Support:

In BIOS settings under PCIe configuration, ensure Above 4G Decoding is set to Enabled and ReBAR Support is set to Auto / Enabled.

If crashes persist on older motherboard chipsets, test with ReBAR disabled.

Update Motherboard UEFI/AGESA Firmware & Adjust CPU VDDCR/SoC Voltages

Solution:

Root Cause: Outdated PCIe Controller Microcode & CPU Memory Controller Instability

High-end GPUs push the CPU's integrated PCIe root complex and System-on-Chip (SoC) memory controller to maximum throughput limit. Outdated AGESA (AMD) or microcode (Intel) firmware causes handshake timeouts between the CPU PCIe lanes and the GPU driver stack.

Step-by-Step Fix:

1.Update Motherboard BIOS/UEFI Firmware:

Visit your motherboard manufacturer's official support page (ASUS, MSI, Gigabyte, ASRock).

Download the latest stable BIOS update containing updated AGESA/Intel microcode for PCIe 5.0 stability.

Flash the BIOS using USB Flashback or built-in EZ Flash utilities in the UEFI menu.

2.Stabilize CPU SoC / VDDCR Voltages:

On AMD AM5 platforms, ensure VSOC Voltage is manually capped between 1.20V and 1.25V to prevent memory controller instability without overvolting.

On Intel LGA1700/1851 platforms, verify CPU VDDQ and System Agent (VCCSA) voltages are operating within nominal limits.

3.Run System File Checker (SFC) and DISM Repairs:

Open PowerShell as Administrator and run the following system repair commands:

powershell
     dism /online /cleanup-image /restorehealth
     sfc /scannow

In which game engines or graphics APIs does the crash repeatedly trigger?

DirectX 12 / Unreal Engine 5 games crash with DXGI_ERROR_DEVICE_HUNG or Frame Generation active.
Vulkan API or heavy Ray Tracing/Path Tracing titles crash upon entering the game engine world.

Disable Optical Flow / Frame Generation Overlays & Limit Rendering Pipelines

Solution:

Root Cause: Optical Flow Accelerator Queue Overflow & DX12 Pipeline State Mismatch

DLSS 3.5 / 4 Frame Generation and FSR 3 / 4 Fluid Motion Frames utilize hardware optical flow accelerators to generate interpolated frames. Intersecting third-party overlays (RivaTuner Statistics Server, Discord Overlay, GeForce Experience, Steam Overlay) hooked into the DirectX 12 swapchain cause frame latency queue overflows, triggering DXGI_ERROR_DEVICE_REMOVED.

Step-by-Step Fix:

1.Disable In-Game Frame Generation Overlays:

Open the game graphics options and temporarily disable DLSS Frame Generation or AMD Fluid Motion Frames (FMF).

Test game stability under standard rendering modes.

2.Disable In-Game Overlays:

Turn off Steam Overlay: Steam Settings > In-Game > Disable Enable the Steam Overlay while in-game.

Disable Discord In-Game Overlay and GeForce Experience / AMD Adrenalin overlay hotkeys.

Close background monitoring programs like RivaTuner / MSI Afterburner overlay.

3.Delete Game Configuration Engine Files:

Open %LOCALAPPDATA% in Windows File Explorer.

Locate the specific game folder and rename or delete Saved\Config\WindowsNoEditor to reset engine settings to defaults.

Reset Vulkan API Runtime Layers & Re-verify Steam Game Files

Solution:

Root Cause: Vulkan Explicit Validation Layer Corruption & Damaged Engine Assets

Vulkan games rely on implicit and explicit API layer registration inside the Windows Registry. Corrupted third-party layer hooks (such as OBS Vulkan capture or old screen recorders) cause illegal memory execution calls when ray tracing BVH (Bounding Volume Hierarchy) structures are allocated.

Step-by-Step Fix:

1.Clean Vulkan Registry Implicit Layers:

Press Win + R, type regedit, and navigate to:

HKLM\SOFTWARE\Khronos\Vulkan\ImplicitLayers

Delete obsolete registry entries referencing non-existent software paths or old video capture tools.

2.Verify Integrity of Game Files:

In Steam: Right-click game > Properties > Installed Files > Verify integrity of game files.

In Epic Games Launcher: Click the three dots on the game tile > Manage > Verify.

3.Disable Hardware Ray Tracing BVH Acceleration:

In game settings, lower Ray Tracing settings from High/Path Tracing to Low/Off to determine if BVH ray traversal hardware logic is triggering driver level timeouts.

Related Diagnostic Guides

Ubuntu Black Screen After Nvidia Driver Install - Secure Boot Diagnostic

SSH Connection Refused on Port 22 But Service Running - Firewall Tree

Linux Boot Stuck at GRUB Rescue After Windows Update - Dual Boot Fix

SmartScreen Can't Be Reached Right Now Error Fix - Windows Diagnostic Guide

Account Locked After Too Many Attempts in Microsoft 365 - Admin Unlock Tree

Subscribe to StepDiag

Embed Guide on Your Website