My Radeon 7900 XTX worked normally for nearly three years, but it began crashing last November and has gradually gotten worse. Now every game renders roughly one frame before the entire computer hard-resets and boots back into Windows.
FurMark is able to run for 18 minutes at full power without any issues, so the card can sustain around 400 watts under a steady load. The crashes only happen when a game actually starts, and the GPU is around 63°C when the reset occurs, so overheating doesn't seem likely.
I've already replaced the power supply and PCIe riser cable, tested both Gen 3 and Gen 4 settings in BIOS, performed several DDU driver clean installs, reinstalled Windows, run SFC and DISM, and kept the RAM at stock settings. The four storage drives report no problems, and I repasted the GPU previously. The system has a Ryzen 9 7900X3D, 32 GB of RAM, an MSI X670E Carbon motherboard, a 1200 W PSU, and Windows 11.
Does a system that passes FurMark but hard-resets on the first frame of a game point to a particular fault? I'm planning to test the graphics card in another computer, but I'd appreciate any other ideas first.
1 Answer
This pattern can still point to a power-delivery problem. FurMark applies a fairly steady load, while games constantly change GPU power draw and can create sharp transient spikes. A PSU may handle the sustained FurMark load but trip during those rapid changes, especially if it has degraded or if the GPU power leads are sharing an unsuitable rail.
Since you've already replaced the PSU, the next useful step is testing the card in a completely different system. If the same instant-reset behavior follows the 7900 XTX, the card is the leading suspect. If it works there, investigate the motherboard, PCIe slot, power cabling, or platform-level stability. On a multi-rail PSU, also make sure the GPU's connectors are distributed according to the PSU manual rather than overloaded on one rail.
I suspected the PSU too, but the replacement is only about six months old. It seemed to solve the problem for roughly two months, then the crashes returned and have become much more frequent over the last couple of weeks.