When I play graphically demanding games, the display suddenly cuts out after a few seconds of heavy GPU activity. The fans ramp up noticeably, but I can still hear game audio and Discord, so the system itself seems to remain running. I have tried using DDU and installing both older and current graphics drivers, reseating the GPU and RAM, and testing the PSU. The issue is reduced or completely avoided when I undervolt the GPU or limit its power to 70%. I also tested integrated graphics with OCCT and did not reproduce the crash. My system has an Intel i5-13600K, RTX 3070, and 750W Gold-rated PSU. Does this sound more like a failing GPU, a power-delivery problem, overheating, or something else?
3 Answers
The fact that audio continues means the operating system and much of the system are still functioning; the graphics driver or GPU has probably stopped responding. A PSU problem is not ruled out just because the rest of the PC stays powered. A brief power-delivery fault can make the GPU lose output without immediately rebooting the whole machine, and a basic PSU stress test may not reproduce the transient load from a real game. Check the GPU power connectors, avoid daisy-chained PCIe cables if possible, and test with another PSU or GPU if you can.
Check GPU temperature, hotspot temperature, and memory temperature rather than only the normal core temperature. An aging card can have dried thermal paste or deteriorated thermal pads and become unstable under sustained load. Limiting FPS with V-Sync is another useful test because it reduces power draw. Also try default RAM settings, reseat both memory modules, and test with another display cable or monitor connection just to rule out simpler causes.
The GPU is the most likely culprit. A graphics card can lose display output under load without showing obvious artifacts first, especially if lowering the voltage or power limit makes the problem stop. That points toward the card becoming unstable when it draws more power or reaches higher boost clocks. If possible, test the RTX 3070 in another system or test another known-good GPU in this one.
That was my suspicion too. I usually associate a dying GPU with visual artifacts, so I wasn't sure whether a display-only crash could indicate the same thing.
The GPU temperatures looked normal during testing, but I'll check the hotspot and memory readings as well. I'll also try a different cable and run the RAM at default settings.