Passmark experienced a crash involving the 7950X3D device.
Passmark experienced a crash involving the 7950X3D device.
I've just moved to a 7950X3D (previously 5900X). Benchmarks and stress tests like CBR23, OCCT, and 3DMark all passed. Surprisingly, the PC restarted in CPU testing within one second after launching the AVX test. No warnings in the Event logger, no WHEA errors, just an error log during reboot indicating a shutdown failure. It seems overheating might be the cause—previously it never dropped below 85°C and stayed around 90°C without issues. The TR PA120 appears fine for this model, and the fans were upgraded to Arctic P120 Max. I’m not sure about other cooling tweaks beyond that. Upgrading the cooler seems like the next step, and the chip is already undervolted (-30 on CCD0 with cache, -20 on CCD1). Any suggestions?
The cooler won't suffice. It seems the 7950X wouldn't have been sufficient, especially with X3D. The issue might lie in reaching 100°C quickly or an unstable VF table.
Perhaps temperatures are acceptable but excessive undervolting might be causing issues, though I question it given the 7950X3D likely using top-grade silicon. If it's about temperature, the P600S case could be restricting airflow, allowing the PA120 to absorb much of the GPU heat output. The 7900XT is a power-hungry machine, along with other factors like levent discussed earlier.
It seems the system might simply power down and restart. RAM performance appears inconsistent. You may want to adjust frequency, timings, and voltage manually instead of relying on EFI settings. This often resolves the issue. If not, test with auto mode to see if crashes persist. Also verify stability under default PBO configurations.
Black screen on reboot; using Expo on GSkill Trident Z Neo / Hynix M die 6000CL30. MCR turned off, training time about 30 seconds. GPU idling on CPU test at low temps below 40°C. Removed front and top steel panels to improve airflow. If voltage drop occurred, expect WHEA error in Event log. Previously, undervolting too much caused issues with TR PA120. Cost was around $30. Consider reinstalling a 280 Eisbaer AIO or getting a new model like Freezer III or Icefloe (good value). Thanks for the advice!
Test with ram at auto speeds. (just when testing) If no crashing, then ram is not stable at expo.
It's no surprise the temperatures were poor; my rear fan was installed incorrectly when I needed to take it out for a swap. Now the temps are about 3°C better everywhere. Passmark SSE tests passed successfully, though it occasionally crashes or restarts—possibly due to a brief desk break. Definitely not the high temps under 80°C. The undervolt seems too aggressive and I might adjust the LLC again. If it keeps crashing, I'll run it with EXPO off.
Final update: crash originated from unstable undervolt. CCD1 was at -30 with a 3D cached unit. Functioned perfectly in testing but not in this case. Reduced to -20 and passed the benchmark.