Claude blamed my RTX 5080’s VRAM for crashes, but the real culprit was far more surprising

Claude, a chatbot, indicated that the VRAM of the user's RTX 5080 graphics card was causing game crashes, but the actual source of the problem was more surprising. The user had been experiencing crashes in Escape From Tarkov, and Claude analyzed the game's detailed crash logs, which included a "DXGI_ERROR_DEVICE_REMOVED" error. This error occurs when the GPU stops responding and the driver has to reset it, leading to game crashes or even system crashes.

Claude guided the user through PowerShell commands to examine Windows event history, which revealed a series of NVIDIA driver resets and kernel watchdog dumps closely followed by game crashes. This led Claude to diagnose the issue as a VRAM defect in the user's graphics card.

Curious about why the card suddenly started failing, Claude suggested running a GPU memory test using memtest_vulkan, a free tool that checks video memory integrity. The first run of the test caught a single flipped bit, which Claude interpreted as evidence of a VRAM defect, prompting the user to initiate an RMA process with the manufacturer, MSI. However, the user questioned this diagnosis due to the card's history of stability for several months.

Further investigation into the test results revealed a significant discrepancy between the VRAM test results and the manufacturer's rated performance. The user's RTX 5080 was capable of a throughput of around 850GB/s, far below its rated 960GB/s, suggesting a potential issue with the VRAM. The user also discovered that the memory overclock was set higher than the Ryzen 7 7800X3D's rated speed for DDR5-6000 memory, which could have contributed to the instability.

After running a long overnight memory test, the user found 22 errors, confirming the VRAM as the root cause of the crashes. To resolve the issue, the user removed two RAM sticks, stabilizing the system for Escape From Tarkov and other games. While the user appreciated Claude's ability to quickly parse through logs and identify relevant errors, they relied on their own expertise for testing, emphasizing the importance of ruling out other components before acting on AI diagnosis.