My ASUS TUF RTX 3080 Ti Gaming OC has started causing DPC_WATCHDOG_VIOLATION blue screens and complete freezes, mostly when launching games. LatencyMon reports nvlddmkm.sys latency around 1298 microseconds. I tested an RTX 2080 in the same system using the same Windows installation, driver, motherboard, power supply, and PCIe slot. With the 2080 installed, latency dropped to about 299 microseconds and LatencyMon reported no problems.
The system uses a Ryzen 7 9800X3D and runs games at 1440p. Under load, the 3080 Ti reaches about 71°C core, 82°C hotspot, and 92°C memory junction, with a peak board draw around 342W. The 12V rail stays near 11.97V or higher. Its clocks and voltage also appear normal, boosting to roughly 1965 MHz at 1.056–1.081V, while idle power states transition normally.
I originally noticed the issue after updating the NVIDIA driver, but a DDU cleanup and rollback from driver 616.56 to 591.74 did not help. The card had previously reached a 98°C hotspot, although I repasted it and temperatures have been reasonable since then.
Event Viewer records WHEA-Logger Event ID 1 and BugCheck Event ID 1001 around the crashes. The stop code is 0x133, with several different parameters. WinDbg identifies nt!KeAccumulateTicks and a recurring nt!ExpGetPoolTagInfoTarget routine rather than naming a specific driver. Does the comparison with the RTX 2080 point to a failing or aging 3080 Ti, or could this still be a Windows, driver, or software problem?
4 Answers
The normal temperatures, clocks, and 12V readings do not rule out a GPU or VRAM fault. They mainly show that the card is not obviously overheating or losing power. WHEA Event ID 1 without useful MCA data is also not enough to identify the exact component, so the strongest practical test is still swapping hardware and checking whether the failure follows the card.
Do not rely on BlueScreenView or WhoCrashed when diagnosing this. Seeing ntoskrnl.exe usually only means the kernel was where the system finally stopped, not that the kernel caused the problem. WinDbg is the better tool, and your repeated DPC watchdog stacks plus the GPU-specific latency difference make the 3080 Ti path worth investigating.
Before declaring the card dead, test it with a conservative configuration: remove any overclock or undervolt, try a substantial core and memory underclock, and test with hardware-accelerated GPU scheduling and overlays disabled. Also try a known-stable older driver or a clean supported Windows installation. If the crashes and excessive latency disappear only when the 3080 Ti is underclocked—or remain unique to that card after those tests—it is probably deteriorating hardware, even though temperatures and voltage readings look normal.
The fact that the problem follows the 3080 Ti while the motherboard, PSU, slot, Windows install, and driver remain the same is strong evidence against a general Windows or platform issue. It does not prove the GPU is physically dying, but the card, its VRAM, or its driver interaction is now the leading suspect. A failing GPU can produce watchdog timeouts without the crash dump naming nvlddmkm.sys directly.
That was my concern too. The lower latency with the 2080 is the main reason I started suspecting the 3080 Ti rather than the rest of the system.