Black screen, high fan speed, GPU stops working?

If the NVIDIA “clean installation” did not use DDU in Safe Mode, completely remove the driver with DDU and test a slightly older NVIDIA driver before changing anything else.

1. Perform a controlled driver rollback

A similar Windows 11 system with an RTX 4070 Ti Super and gaming crashes was advised to remove the NVIDIA driver completely and test older driver versions ^1^.

1. Download DDU (Display Driver Uninstaller) from its developer and download a slightly older RTX 4070 Ti Super driver directly from NVIDIA.
2. Disconnect from the internet temporarily so Windows Update does not install another display driver during the procedure.
3. Enter Safe Mode:
   • Settings → System → Recovery → Advanced startup → Restart now
   • Troubleshoot → Advanced options → Startup Settings → Restart
   • Select Safe Mode.
4. Run DDU and choose GPU → NVIDIA → Clean and restart.
5. Back in normal Windows, install the older NVIDIA driver using Custom installation and install only the display driver and PhysX.
6. Restart, reconnect to the internet, and test the same game with the same settings.

Treat this as a diagnostic rollback, not proof that the current driver is defective. If the older driver fails identically, move to evidence collection and hardware isolation rather than repeatedly reinstalling drivers.

2. Determine whether Windows recorded a GPU timeout

Your load-triggered signal loss and fan ramp-up closely resemble a graphics Timeout Detection and Recovery (TDR) failure: Windows detects that the GPU stopped responding and attempts to reset it. A failed reset can produce VIDEO_TDR_FAILURE (0x116), but the underlying cause can still be a driver, cooling, power delivery, memory configuration, motherboard, or defective component ^2^.

After the next incident:

1. Search for Reliability Monitor and open View reliability history.
2. Open the critical event at the failure time and record:
   • LiveKernelEvent code, particularly 141, 117, or 1b8
   • Bug-check code
   • Bucket ID and referenced files
3. Open Event Viewer → Windows Logs → System.
4. Filter to the failure time and look for display-driver, WHEA, Kernel-Power, or bug-check events. Kernel-Power after a forced shutdown records the shutdown; it does not by itself identify the original fault.

Check these dump locations:

C:\Windows\Minidump
C:\Windows\LiveKernelReports
C:\Windows\LiveKernelReports\WATCHDOG

Windows normally stores full live dumps under LiveKernelReports and component-specific minidumps in its subdirectories ^3^.

3. Extract the useful dump fields

Install WinDbg with:

winget install Microsoft.WinDbg

WinDbg supports Windows 11 and can analyze crash dumps ^4^. Open the newest dump, then run:

.symfix
.reload
!analyze -v
lmvm nvlddmkm

Record only:

• Bug-check or live-dump code and four parameters
• MODULE_NAME
• IMAGE_NAME
• FAILURE_BUCKET_ID
• The first meaningful stack entries

nvlddmkm.sys or dxgkrnl.sys identifies the graphics path involved, but does not by itself distinguish a faulty NVIDIA driver from unstable GPU, PSU, motherboard, or RAM hardware. DirectX dumps can preserve GPU and driver state from the moment of a TDR, although available detail varies by driver ^5^.

Crash dumps and debugger output can contain usernames, paths, process information, memory contents, or credentials. Prefer posting only the sanitized fields above. If a technician needs the dump, use a restricted OneDrive link shared only with that technician, review it for identifying data, and revoke access afterward—do not publish it openly.

4. If the rollback fails, isolate hardware under load

Because you already replaced the display cables and monitors, do not repeat those checks.

1. Return BIOS and tuning to defaults. Disable CPU/GPU overclocking, undervolting, and memory overclock profiles temporarily. Incorrect compatibility or memory timings and overclocked components are documented TDR causes ^2^.
2. Log sensors during the same game. Capture GPU temperature, hotspot temperature, GPU power, fan speed, and CPU temperature until failure. Do not infer overheating from fan speed alone.
3. Inspect GPU power and seating safely. Shut Windows down, switch the PSU off, unplug it, and wait before touching components. Reseat the GPU and its power connector; ensure the connector is fully inserted and not sharply bent at the plug. Never open the PSU.
4. Run Windows Memory Diagnostic: search Windows Memory Diagnostic → Restart now and check for problems. After reboot, inspect Event Viewer → Windows Logs → System → MemoryDiagnostics-Results ^6^.
5. Use substitution testing. The decisive tests are this GPU in another known-good PC, another suitable GPU in this PC, or a known-good adequately rated PSU. If only GPU-heavy workloads fail while CPU and memory diagnostics pass, prioritize GPU and PSU substitution.

If the vendor could not reproduce it, provide the exact game, graphics settings, approximate time-to-failure, Reliability Monitor code, sanitized WinDbg fields, and sensor log. That gives the repair shop a reproducible workload and evidence instead of a generic “black screen” report.

References

1. New build Windows 11 repeated, random restarts - Microsoft Q&A
2. Bug Check 0x116: VIDEO_TDR_FAILURE
3. Kernel Live Dump Code Reference
4. Install the Windows debugger
5. DirectX Dump Files
6. Bug checks (stop code errors)