katago hangs and causes windows BSOD

[peterwang]

Description

I am using katago-v1.15.3-trt10.2.0-cuda12.5-windows-x64 and weights file kata1-b28c512nbt-s8209287936-d4596492266.bin.gz.

While it is analyzing an sgf file (normal game file nothing special), it hangs and Windows does not response any inputs. After about 2 minutes, windows just pops a blue screen, saying it is going to reboot (which I guess is caused by high temperature of GPU, since the fans is noisy and laptop is very hot).

BTW, the weights file kata1-b28c512nbt-s8032072448-d4548958859.bin.gz has been OK by now.

Does this mean that the issue is in the weights? How can I debug such issue?

Hardware and driver info

GPU 0: Quadro T2000 with Max-Q Design Driver Version: 555.85 CUDA Version: 12.5

Software info

katago-v1.15.3-trt10.2.0-cuda12.5-windows-x64.zip TensorRT-10.2.0.19.Windows.win10.cuda-12.5.zip cuda_12.5.0_555.85_windows.exe

Thanks.

Update

BTW, the weights file kata1-b28c512nbt-s8032072448-d4548958859.bin.gz has been OK by now. This also caused BSOD too.

Update 2

After rolling back the nvidia driver and using the katago without tensorrt / cuda support, BSOD problem goes away. Seems the culprit is the driver!

[FengmingGo]

Try to reduce the number of threads？