serizba / cppflow

Run TensorFlow models in C++ without installation and without Bazel
https://serizba.github.io/cppflow/
MIT License
787 stars 179 forks source link

Adjusting GPU memory usage doesn't work #242

Open pgeflip opened 1 year ago

pgeflip commented 1 year ago

Hi everyone, I am currently trying to get cppflow to work together with tensorflow-gpu-2.10 for Windows.

I am currently encountering the error "failed to create cublas handle: CUBLAS_STATUS_NOT_INITIALIZED" during a simple inference attempt.

A simple Google search attempt indicates insufficient video memory. This also sounds logical. So I tried https://gitlab.ipr.kit.edu/usjsb/cppflow/-/blob/master/docs/source/quickstart.rst aswell as https://github.com/serizba/cppflow/issues/102, but it does not seem to have an effect. Still the same portion of VRAM allocated for the tensorflow session (1249 MB) as without these changes. My system has a NVIDIA Geforce MX450 GPU with 2GB VRAM.

Does anyone have any clue?