New user GPU reset with NVIDIA 610.88
Collection window: 17 September 2026, 01:20:00–01:32:29.3719757 CEST. System/Application events are preserved as complete event XML in JSON and as native EVTX exports. Reopening the EVTX files confirmed 11 System and 53 Application records. Player logs were preserved separately by the primary agent and were not modified by this collection.
Classification: a new NVIDIA GPU timeout/reset, with Windows LiveKernelEvent 117. The first OS event is 2026-09-17T01:29:52.3631892+02:00 / 2026-09-16T23:29:52.3631892Z. The ordinary formatted NVIDIA message was unavailable, so the following payloads come from the retained event XML:
| Local CEST time | System record / event | Relevant payload |
|---|---|---|
| 01:29:52.3631892 | 202586 / nvlddmkm 14 | Device Video12, diagnostic words beginning 024c1547. No undocumented interpretation of these words. |
| 01:29:52.3631892 | 202587 / nvlddmkm 153 | UCodeReset TDR occurred on GPUID:100 |
| 01:29:53.6776682 | 202588 / nvlddmkm 153 | Resetting TDR |
| 01:29:54.4710482 | 202589 / nvlddmkm 153 | Reset TDR |
| 01:29:54.4924177 | 202590 / nvlddmkm 153 | Restarting TDR |
WER records 284217 at 01:30:03.537738 and 284268 at 01:30:06.3344736 identify the same new LiveKernelEvent, code 117, and WATCHDOG-20260917-0129.dmp. They are two publications of one report, not evidence of two crashes. The other 50 WER records published in this window reference earlier incidents; their fresh publication times must not be used as new crash times. In particular, the republished 193/133 reports are older incidents. There are no new Kernel-Power, WHEA, Application Error or Application Hang events in this query window. This identifies a GPU reset, not a new Windows reboot or a proven managed exception.
Active driver evidence goes beyond the installer result. The read-only query at 23:32:45.7575078Z reports:
nvidia-smi --version: KMD 610.88, NVML 610.88; GPU query: driver_version 610.88, RTX 4060 Laptop, exact subsystem0x299D1043; both commands exit 0.- PnP: 32.0.16.1088,
oem227.inf. The running/startednvlddmkmservice points tonvami.inf_amd64_46b14a8ba26be7f7/nvlddmkm.sys; that on-disk image is 32.0.16.1088, SHA-256C28FD719733F0AF2E56D48EF0EF1ABE67E6FA71972D870C2F03D2D66FCC509F0. An on-disk version alone would not prove the live kernel version; the independent KMD query addresses that limitation. - Boot time remains 16 September 22:40:30.500 CEST. UserPnp recorded successful addition of
nvlddmkmat 01:23:53.8889219, before the new failure. No Windows restart occurred for this comparison. - The primary agent independently confirms Player.log initializes 32.0.16.1088 / D3D11, build GUID
863b7c4dc60d4b53971aa9bc445b51e2. It records three alps-02 starts, Low / Rain / seed 9909140, from the normal current candidate EXE without Forward or isolation switches. This is a manual user test, not a matched repetition of the earlier automated Forward baseline.
The separately preserved incident index has a pre-reset entry at 23:29:49.2849886Z (3.078 seconds before the first OS event, third-race trigger time 64.899 s) and further entries after the reset. These are gameplay incident records, not GPU render-completion samples or a clean-exit proof. No exact failure-at-lap-transition claim is made from them.
The new report queue directory was created at 23:30:03.2459838Z. Its protected Report.wer could not be read without elevation. Access to the LiveKernelReports root was denied; direct metadata lookups of the named WATCHDOG and SystemTemp files returned not found. Therefore their contents/continued physical availability are not established by this collection, although their paths are recorded by WER. No protected dump was copied, no new UAC requested and no native dump stack analyzed.
The observed manual configuration still resets under active 610.88; changing to that driver did not repair this user test. This does not identify a hardware defect or prove an exclusive engine/driver cause. Installer Finish and optional component changes were not independently observed and are not claimed.
A subsequent read-only NVAPI before/after comparison confirms all four previously recorded overrides unchanged in both Base and CurrentGlobal: refresh override 0, power state 1 (Prefer Maximum Performance), SHIM_MCCOMPAT 1 and SHIM_RENDERING_MODE 1. The profile lookup summary found no matching profile for the queried Rampage executable names/paths. This verifies those specific settings and lookups; it does not guarantee preservation of every driver setting, NVIDIA App option or optional installed component.
Evidence: active driver and boot, TDR timeline, WER index, complete events, EVTX hashes/counts, protected-artifact access results. This collection performed no installation, UI action, rendering workload, reboot, registry/profile change or elevation.