A friend's reference-model AMD Radeon RX 6750 XT crashes consistently during gaming, despite passing FurMark and Prime95 stress tests for more than 30 minutes. During gameplay, the image briefly stutters, the GPU core clock suddenly drops to 0 MHz, the screen freezes, and Windows shows a blue-screen error before rebooting. The problem only appears in real games, not synthetic torture tests. The system uses a Ryzen 5 7500F, MSI B650M Gaming motherboard, Seiwhale DDR5-4800 memory, an Xuanwu 850W 80+ Bronze power supply, Windows 10 22H2, and Adrenalin driver 26.3.1. Could transient power spikes, the power supply, graphics drivers, or memory instability explain this behavior, and what should be checked next?
3 Answers
Run a proper memory test as another elimination step. DDR5-4800 is not an aggressive speed for this platform, but unstable or defective memory can still cause game-only crashes. Test with MemTest86 for several passes or overnight, and try one memory module at a time if errors appear. You can also temporarily remove any GPU undervolt, overclock, custom tuning, or unusual driver settings and perform a clean graphics-driver installation. If the issue continues with verified memory, a known-good PSU, and clean drivers, the graphics card itself becomes a stronger suspect.
Collect the Windows crash dumps before changing too many things. Check the Windows Minidump folder for files created around the crashes and review them with a dump-analysis tool. The stop code and the driver or module named in the dump can help distinguish a graphics driver failure from a hardware, memory, or power-related reset. If no dumps are being created, enable Small Memory Dump generation in the Windows recovery and startup settings, then reproduce the crash and compare several results.
The first thing I’d investigate is the power supply. A steady FurMark load does not necessarily reproduce the sharp, rapidly changing power demands that occur in games. An unfamiliar 850W unit can still struggle with transient spikes even if its rated wattage looks sufficient. Since the GPU clock dropping straight to 0 MHz before the freeze resembles a hardware protection event, test the system with a reputable, known-good PSU if possible. Also make sure the graphics card uses separate PCIe power cables rather than one daisy-chained cable. The replacement PSU should be checked as well, since simply swapping to another unit does not confirm that it is suitable.
The PSU has already been swapped, but the replacement model’s quality and transient-handling capability still need to be verified. The RAM is also from a lesser-known brand, so it would be worth testing it separately rather than assuming the swap ruled out power problems.
Several dump files would be especially useful because a single crash report can be misleading. It is also worth recording the exact stop code and whether every crash points to the same graphics driver component.