Deep learning inference with Windows UWF actived

AI & Data Science Deep Learning (Training & Inference) cuDNN
1

Hello,
I’ve a Windows application the does deep learning inferences on a NVidia 1660 with Cuda 9.0 and CuDnn 7.x using tensorflow framework.
I’ve enabled the windows UWF filter (https://docs.microsoft.com/en-us/windows-hardware/customize/enterprise/unified-write-filter) to protect the file system against sudden shutdown.
I’ve installated the GPU a month ago to speed up the system and before, when it ran in CPU, all worked fine. Now in two days the system fills all the overlay memory of UWF (2048MB).
Does CUDA write somithing (logs, temporany files) in hard drive when I use it? I’ve already unlocked the folder in %appdata%/NVIDIA/ComputeCache, but it didn’t work.

2

Hi,

It seems there might be some cache files generated at %APPDATA%NVIDIA\ComputeCache if you are using an old version of driver.
Can you try updating the driver to latest version?

Thanks

3

Hello,
thanks for your reply. I’ve already added the folder %APPDATA%NVIDIA\ComputeCache to UWF exceptions. Does it write only this cache?

4

No we don’t write anything else unless you turn on API log.
Please refer below link:

Can you try updating to latest CUDA/cuDNN version?

Thanks