HOWTO: Install Tensorflow on FreeBSD

However I am puzzled about one thing. What is the purpose of running Tensorflow on FreeBSD? I was under impression that NVidia is not releasing binary blob CUDA GPU drivers for FreeBSD. Does NVidia support GPU CUDA computing on FreeBSD? If the answer is NO than compiling Tensorflow on FreeBSD is an exercise in masochism.
Well at least I hope AMD GPUs works with tensorflow on FreeBSD... This is the reason I would rather not rely on frameworks which doesn't use OpenCL. I do think Nvidia CUDA might have FreeBSD support. However you can use Nvidia/AMD or any GPUs with OpenCL so long the GPU vendor supports OpenCL standards.

"OpenPose" seems to have OpenCL support. I'm trying to get caffe installed since it is required for OpenPose.
Also "mediapipe" uses OpenGL ES which should also work on FreeBSD if you know how to install everything, I have issues with bazel to build c++ code.
 
Tensorflow 2.9.1 is in the port tree with the latest keras. Tensorflow also builds with opencl, but I don't have a amd or intel card to test. as for Cuda. It could possibly work, with a wrapper.

Thanks. My AMD RX-580 GPU supports only OpenCL 1.1, Tensorflow only support OpenCL 1.2, not sure if you mean wrappers for backwards compatibility with OpenCL version which I read is possible to do so.

Also it seems Tensorflow and MediaPipe frameworks uses "bazel" for installation and building projects, I do not like to touch these new popular bloatware riddled into these popular frameworks, so I had to ditch both of them since "bazel" was giving me a headache...

Thankfully google showed a framework called "dlib" which uses "cmake" and C++11 and it works better than OpenCV in face detection and face recognition from the examples given by "dlib", all working good on FreeBSD 13.1

I'm also trying to figure out how to compile "caffe", it seems many FreeBSD users had issues installing it.
 
Back
Top