GPU no-signal / GPU lost in Minecraft and OCCT; improved after AMD chipset update (causation unconfirmed)

Please investigate or route this report to the display-driver/desktop GPU team. This is a family-owned custom PC with repeatable failures before the chipset update, now passing initial validation. We are not asserting a confirmed NVIDIA or AMD defect, nor ruling out intermittent hardware/power/PCIe problems.

SYSTEM
ASUS NVIDIA GeForce RTX 4070 Ti SUPER, 16 GB; exact ASUS SKU unverified. VBIOS 95.02.6c.00.77. Ryzen 9 7950X3D; MSI MAG X670E TOMAHAWK WIFI (MS-7E12), BIOS 1.D0 dated June 18, 2024, unchanged during testing. 32 GB G.Skill F5-6000J3238F16G (2x16), configured 4800 MT/s. Windows 11 Pro 25H2 build 26200.9457. WD_BLACK SN850X 2 TB. Samsung monitor (exact model/firmware/cable length not yet known); current WMI output technology 5 (HDMI), directly connected to GeForce during latest passing tests. PSU model and wiring have not been verified. No manual clock changes during investigation.

REPRODUCTION BEFORE MITIGATION
Launch Minecraft Java Edition 26.3 from Minecraft Launcher with bundled Java 25.0.1. Observed settings: 260 FPS cap, render and simulation distance 32, VSync off. Play for several minutes. Physical display loses signal while fans/RGB remain on; the PC can remain accessible over SSH. Vanilla and modded Minecraft have both been used; the Oct 6 monitored failure was vanilla. NVIDIA overlay was disabled and the failure persisted. Other overlays/background programs, including Crosshair X, were not fully isolated. Safe Mode had remained visible for at least 15 minutes. Minecraft settings were not normalized to a minimal-load profile.

FAILURE TIMELINE (all times EDT, UTC-4, October 6, 2026)
20:07, driver 591.86 / 32.0.15.9186: physical no-signal confirmed. NVIDIA OpenGL event 1, error code 3/subcode 2; javaw Application Error 1000 in nvoglv64.dll, exception 0xc0000409, offset 0x10907cd. nvidia-smi reported GPU lost (exit 15). Last valid GPU core sample 47 C / 131 W; session peak 54 C. Reboot recovered video.
21:13–21:18: official signed NVIDIA 617.14 installer, clean installation and reboot; verified Windows driver 32.0.16.1714.
21:31:44–45: same Minecraft/no-signal failure on 617.14, NVIDIA OpenGL error code 3/subcode 2, javaw crash in nvoglv64.dll 32.0.16.1714, 0xc0000409, offset 0x10af33d. Last valid core sample 49 C / 152.93 W; session peak 56 C. GPU lost; SSH still responsive. A subsequent reboot recovered video.
21:45–21:50: OCCT Personal 17.1.7, 3D Adaptive Variable, discrete GPU, 20–100% load, +5 percentage points every 30 seconds, planned 15 minutes. Failed about 5 minutes in: physical no-signal, NVIDIA telemetry stopped responding. Last valid sample 61 C / 240.63 W; session peak 62 C. Test was stopped, not followed by VRAM testing. A normal reboot did not immediately restore NVIDIA enumeration; a later manual restart did.
Additional collected System event 1001 records report bugchecks 0x113 and 0x133 at 21:35 and 21:52 respectively, associated with retained minidumps. These have not been analyzed for driver attribution. Historical matching Java/nvoglv64 events were found back to September 21 within a bounded 45-day query.

INTERVENTION AND RESULTS
AMD chipset package 8.08.12.551 installed October 6 around 22:24, reboot 22:27. Six installer components reported success:
GPIO Promontory 3.0.3.0 -> 3.0.5.0; PPM provisioning 8.0.0.38 -> 8.0.0.65; 3D V-Cache optimizer 1.0.0.9 -> 1.0.0.12; PSP 5.32.0.0 -> 5.46.0.0; SMBus 5.12.0.38 -> 5.12.0.44; GPIO 2.2.0.134 -> 2.2.0.137.
October 7: identical 15-minute OCCT test passed with motherboard/iGPU display (core peak 65 C, 294.78 W), then passed with monitor directly on GeForce (67 C, 292.77 W); OCCT reported no errors and completed the schedule. User confirmed physical picture remained visible for the direct-GeForce test.
October 7 15:44–16:24: 40-minute Minecraft observation with monitor directly on GeForce, 1,168 valid GPU samples, 37–60 C, 0–99% utilization, 3.90–246.99 W. No captured GPU-lost condition or relevant Windows error. User confirmed vanilla and modded versions all worked without crashes. Four Java starts were intentional version changes, not crashes; this was not uninterrupted gameplay. One approximately 13-second telemetry gap was an expected recorder reconnect at its 33-minute stream lifetime, not established GPU loss.
NVIDIA 617.14 remained installed and BIOS was not updated. Reboots and display reconnections are confounders. We have not rolled back the chipset update to reproduce again, tested a replacement PSU/GPU, performed a dedicated VRAM test, or proven a permanent fix. Core temperature alone cannot exclude hotspot, VRAM or connector/power issues.

EVIDENCE
Attached are selected GPU telemetry CSVs and the AMD install summary. Relevant crash minidumps/watchdog dumps, DxDiag, NVIDIA query output, driver inventory and event logs have been preserved privately for follow-up. Raw dumps are not attached because they may contain personal memory data. No credentials, SSH keys, remote-access configuration, personal paths, game worlds/chat or account identifiers are included in these attachments. One chipset internal DLL log was inaccessible, but the install log and success summary were preserved.
Please acknowledge receipt and advise whether the retained watchdog/minidumps or a specific additional diagnostic is needed, and whether this signature is known to involve NVIDIA OpenGL and AM5 chipset/power-management interactions. Temporary diagnostic remote access is being removed following evidence collection.