My Gigabyte AORUS RTX 5090 causes the displays to go black during demanding games while the fans ramp to maximum speed. Game audio and background music continue, but the system becomes unresponsive and requires a forced shutdown. Windows Event Viewer usually records nvlddmkm errors 14 and 153. On some attempts, the system instead crashes with a DPC_WATCHDOG_VIOLATION and creates a minidump.
The problem can be reproduced consistently in Battlefield 6 and Helldivers 2, regardless of graphics settings or whether DLSS/DLAA is enabled. Less demanding games have never triggered it, and the issue has not occurred outside gaming except briefly while generating AI content with ComfyUI more than a year ago. OCCT and FurMark stress tests have run for 30–60 minutes without reproducing the crash.
My system uses a Ryzen 7 9800X3D, Corsair RM1000e PSU with its included 12V GPU cable, 32GB of DDR5-6000 memory, two Samsung M.2 drives, an ASUS B650E motherboard, and the Gigabyte RTX 5090. I have already reinstalled Windows and drivers, tested the memory and drives, reseated the card and power cable, checked for visible damage, updated BIOS and drivers, changed PCIe settings, monitored temperatures and voltages, and tried reducing the GPU power limit by up to 20%.
Before sending the card in for warranty service, I would like to determine whether this points to a defective GPU, a power-delivery problem, or something else. I can provide minidumps and additional hardware-monitoring data if useful.
2 Answers
The minidumps would be the most useful next step for the DPC_WATCHDOG_VIOLATION crashes. Check the Windows Minidump folder and collect several files if available. They may show whether the failure is actually being caused by the NVIDIA driver or by another driver that is blocking the system. If there are no dumps, configure Windows to create small memory dumps before reproducing the crash again.
A transient power-delivery problem is worth investigating. A 5090 can produce very brief power spikes that may not show up in a normal temperature or voltage log, and a synthetic benchmark does not always reproduce the same load behavior as a particular game. The fact that crashes often happen while entering or leaving a match could fit a rapid load change. Make sure the GPU power connector is fully seated, avoid adapters or extensions, and, if possible, test with a known-good higher-capacity ATX 3.x power supply. Since lowering the power limit did not solve it, that does not completely rule out the PSU or the card's power circuitry.
I suspected power delivery too, especially because the crashes initially happened most often while loading into or leaving a match. I tried reducing the GPU power limit, but the issue still occurred, and synthetic stress tests remain stable.