I built a small-form-factor PC and used it for about a month before leaving it powered off for more than a year. After bringing it back into service, I experienced random blue screens, and a few weeks later the NVMe SSD failed and had to be replaced. The current issue is that demanding games such as Crimson Desert, Half-Life: Alyx, and Dyson Sphere Program randomly close to the desktop without an error message, freeze, or blue screen. The crash timing varies: sometimes it happens in the menu, while other times I can play for 10–20 minutes first. Occasionally, Reliability History or Event Viewer records an error. This happens on Windows 11, CachyOS, and Bazzite, but synthetic tests and stress programs—including 3DMark, Superposition, Heaven, CrystalDisk, MemTest64, Prime95, FurMark, OCCT, and Cinebench—complete without errors. The BIOS is updated and running at stock settings. I have tested with and without memory EXPO/XMP, changed PCIe link speeds, reseated the RAM, SSD, cables, and riser, tried a replacement riser cable, and experimented with different PCIe and M.2 generations. The CPU reaches roughly 90°C in the compact Dan A4 case, although it remains stable during stress testing. Could this still indicate a faulty GPU, power-delivery problem, or another hardware issue that only appears under real game workloads?
2 Answers
A CPU temperature around 90°C during a sustained workload is not automatically dangerous, especially in a very small case, but repasting and checking cooler mounting would still be worthwhile. Make sure the heatsink is making even contact and that the fan and case airflow are working correctly. A benchmark passing does not completely rule out a thermal or power issue, since different games load the CPU, GPU, memory, and PCIe bus in different ways.
A clean driver reinstall is worth trying, preferably by removing the existing graphics driver completely and then installing a current stable version. However, since the same behavior occurs across multiple operating systems and fresh driver installations, a driver-only explanation seems less likely. I would next compare the problem with another GPU or power supply if possible, and inspect the crash records for display-driver resets, hardware-error reports, or application faulting modules. Game workloads can expose marginal GPU, PSU, RAM, or PCIe problems that synthetic tests miss.
I’ve already reproduced it on Windows and Linux with clean installations, so I’m also leaning away from a single driver issue. Testing another GPU or PSU may be the most useful next step.
The CPU reached about 91°C during Prime95 and stayed stable, so I was unsure whether repasting would change anything, but I’ll check the mounting and thermal paste anyway.