It crashes due to a video driver failure (nvlddmkm.sys) specifically when the GPU isn't under heavy load.
It crashes due to a video driver failure (nvlddmkm.sys) specifically when the GPU isn't under heavy load.
Hello everyone. I recently acquired a GPU from a hardware swap and have encountered some unexpected BSOD problems. I bought a Zotac 3090 Trinity OC, which performs just as well—or even better—than expected in benchmarks. The main concern arises after running intensive tasks. During AI training sessions, I can run for hours without issues, consuming around 390 watts continuously. However, once the process finishes, the GPU tends to crash within minutes. This behavior is disrupting my plans, especially since I’m preparing to release a new AI suite soon. I’ve tried various fixes: updating BIOS, clearing drivers with DDU, upgrading and downgrading drivers, overclocking, underclocking, updating VBIOS, and adjusting VBIOS settings. Notably, this issue doesn’t affect my 3060ti or 3080 models. While many claim the GPU is failing due to high clock speeds, I’ve observed that performance drops only when the GPU is idle. I’ve recently set its performance preference to maximum, hoping it will prevent crashes and resolve the problem. Thank you all for any suggestions—I’m at a bit of a crossroads. If things worsen, I might dispute the purchase via PayPal, though I’d prefer not to create tension with the seller who has been so supportive.
Hello, yes I confirmed. I previously ran DDU in safe mode and have used it three more times for updates and reverting drivers. The PC is just three days old, with a brand new Windows install and nearly all original components intact.
You’re sure about this? I don’t care if there’s any chance of trouble—my main goal is to avoid anything negative between us. I really believe he didn’t give me a bad GPU because of his effort to assist, not out of malice.
Did this function properly? I’m facing the same problem with the GPU I use for PyTorch workloads in Ubuntu. After training finishes, long runs take hours and only become visible at completion. If I run it for just two hours, the system crashes immediately. Here are the logs you provided: gpu_z_sensor_log_tww.csv, gpu_crash_test.csv, gpu_z_sensor_log.csv, crash_test_2.csv. Updated April 21, 2025 by Thomas2419 including pre-crash logs.