I recently built a PC with a Ryzen 5 7500F, Yeston RX 5700 XT, Gigabyte B840M GAMING PLUS WIFI6E motherboard, 32 GB of second-hand Corsair DDR5-5200 CL40, a Seasonic 750 W modular PSU, and an M.2 NVMe SSD. Windows is stable during normal use, but GPU-heavy workloads can cause a black screen followed by a complete restart. In another benchmark, the entire system froze and required a forced shutdown. The GPU core temperature was only about 60°C, so overheating does not seem obvious.
The RX 5700 XT needs an 8-pin and a 6-pin connector. My PSU only included one PCIe cable, with two 8-pin connectors on the GPU end, so I am currently using that single cable for both GPU connections. I am unsure whether this could cause instability and whether the modular cable connections should be reseated.
The motherboard PCIe x16 setting was initially on Auto. I tried forcing it to Gen 3, but the benchmark still froze at around 31%, so I plan to return it to Auto. I have not changed PCIe bifurcation settings.
I am also considering RAM or platform instability. The memory is used and currently runs at DDR5-5200; I may test it at 4800 MT/s with XMP/EXPO disabled. What should I test first to distinguish between a faulty GPU, power or cabling problem, PCIe or motherboard issue, unstable RAM or CPU, driver problems, and a benchmark-specific failure?
2 Answers
Do not change several settings at once. Start with BIOS defaults, leave PCIe link speed on Auto, disable XMP/EXPO, and run the memory at a conservative setting such as DDR5-4800. Test the RAM with a bootable memory test and, if possible, test each module separately. Used memory can be stable in one system and problematic in another even if it worked previously.
If the system still fails with conservative memory settings, focus on the GPU and its power delivery. Monitor GPU power and hotspot temperature rather than core temperature alone, and verify that every motherboard and GPU power connector is fully inserted. A crash that varies between a reboot and a complete freeze can still come from unstable hardware, so reproduce each test several times instead of relying on one benchmark result.
A black screen followed by a full reboot or hard lock under GPU load points more strongly toward power delivery, the GPU, or general hardware stability than a normal graphics-driver timeout. The 750 W rating should be sufficient, but the cable arrangement and connections still matter. Ideally, use two separate PCIe cables from the PSU, one for each GPU connector. Since only one cable was supplied, carefully reseat both ends of the existing modular cable and inspect the plugs for damage or looseness. Do not use a modular cable from another PSU model.
For isolation, test the GPU by itself with a GPU stress test, then run a CPU-only test. If CPU-only testing is stable but GPU testing consistently causes the failure, try the card in another known-good system if possible. A clean driver installation with DDU can also rule out software, but forcing PCIe Gen 3 not changing the behavior makes a simple PCIe link-speed issue less likely.
The PSU only came with that single PCIe cable, so I am currently using its two GPU-side connectors for the card's 8-pin and 6-pin sockets. I will reseat the modular connections and test the system with separate GPU and CPU workloads before replacing anything.