My gaming PC intermittently loses video output during games and some benchmarks. The TV reports "No Signal," game audio stops, and the system never recovers without holding the power button. Caps Lock and Num Lock still respond, so Windows appears to remain partially active. After selecting Restart, the GPU fans run at full speed for roughly 10–20 seconds before slowing and allowing the reboot to continue.
The problem has occurred with both a Gainward RTX 3080 and my current Inno3D RTX 5070 Ti, making a defective GPU seem less likely. It is most common during shader compilation, heavy asset loading, rapid scene changes, or demanding DX12 workloads rather than simple sustained GPU load. Resident Evil 2 Remake can trigger it around the Mr. X elevator/explosion sequence above 60 FPS, while capping the game at 4K/60 makes that scene reliable. Indiana Jones, Silent Hill Remake, 3DMark Time Spy, and Steel Nomad can also reproduce it.
The system includes a Ryzen 7 5700X, MSI B550M-A PRO motherboard, BIOS E7C96AMS.2G0, 32 GB of Corsair Vengeance LPX DDR4-3200, an Inno3D RTX 5070 Ti using its stock VBIOS and native 12VHPWR cable, an MSI MPG A850GS 850 W PSU, a 2 TB SSD, and a TCL C755 TV connected at 4K/120 Hz over HDMI.
FurMark, Superposition, OCCT VRAM, and an hour-long OCCT memory test pass. The GPU stays around 75°C, PCIe is reported as x16 Gen4, and long sessions in some games are stable. Disabling A-XMP and CPU boost did not solve it. Lowering the GPU power limit to 95% reduces the frequency, but 90% or lower still does not guarantee stability. OCCT 3D Adaptive fails at 4K but passes at 1440p. Event Viewer sometimes records Video_DXGKRNL_LiveDump entries, and a 3DMark log reports DX12 device removal error 0x887A0005 while waiting on a GPU fence.
I have already tried replacing the GPU and PSU, clean-installing current NVIDIA drivers with DDU, disabling XMP and CPU boost, and testing memory and graphics hardware. What should I investigate next: an outdated BIOS, motherboard or PCIe instability, CPU/Infinity Fabric behavior, a driver problem, or a 4K/120 Hz HDMI compatibility issue?
4 Answers
The power-limit behavior is another useful clue, but it does not prove the PSU is bad. Reducing the limit changes transient power and boost behavior, which can hide marginal PCIe, motherboard, firmware, or driver instability. Since the failure is tied to 4K and particular DX12 transitions, test with the TV at 4K/60, then try a different HDMI cable or monitor and temporarily disable HDR, VRR, and other HDMI 2.1 features one at a time. If the problem disappears when using another display or a lower link mode, the TV/HDMI path becomes a strong suspect.
I would update the motherboard BIOS before drawing conclusions. Keep the BIOS completely stock afterward, including JEDEC memory settings, and verify that CPU and DRAM voltage/frequency settings are genuinely stock rather than just left on a board-specific Auto value. Since the issue occurs on a B550 platform, also test PCIe manually at Gen 3 if the firmware provides that option. If Gen 3 is stable while Auto/Gen 4 is not, that points toward a motherboard, slot, signal-integrity, or CPU memory-controller/PCIe issue rather than either graphics card.
The benchmark log shows Steel Nomad completing once and then failing repeatedly at the DX12 fence wait with error 0x887A0005. That makes this more than a random game crash. After updating the BIOS and testing display settings, try a different PCIe slot only as a diagnostic, reseat the card and power connector carefully, and test with the motherboard at default settings. If possible, test the PC with another display connected by DisplayPort. A clean result there would help separate a GPU-rendering problem from HDMI 4K/120 Hz link instability.
The 3DMark message is significant: DX12 is reporting that the graphics device was removed or reset, not simply that the benchmark reached a high temperature. That can come from a driver timeout, PCIe communication problem, unstable firmware, or a display link issue. Check Windows crash reports and any minidumps after a failure, and configure Windows to create small memory dumps if none are being generated. The Video_DXGKRNL_LiveDump entries may contain more useful details than the normal Event Viewer summary.
I already ran DDU offline in Safe Mode and prevented Windows from replacing the driver. I’ll try the BIOS update and stock memory settings, although my current firmware does not appear to expose a PCIe Gen 3 selection.