I recently install caffe2 in tk1. While only CPU mode can be used. Does caffe2 require CUDA 7.5 or higher? How to install it in tk1 with CUDA support?
We test caff2 on TX1 and it works well.
We will check caffe2 for tk1 and update information to you later.
Caffe2 requires NVRTC library which is only available at CUDA v7.0 up.
Jetson TK1 is not supported by the CUDA 7 Toolkit so that can’t build caffe2 with GPU support.
I’ve tried install Caffe2 on TX2(jetpack 3.2.1) from source as in https://caffe2.ai/docs/getting-started.html?platform=tegra&configuration=compile
but there always occured an error of builing onnx-tensorRT(build libonnx). I forked the latest version of Caffe2 and turned on -DUSE_TENSORRT=ON -DCAFFE2_LINK_LOCAL_PROTOBUF=OFF as in
https://github.com/pytorch/pytorch/issues/8167 but nothing helpful.
So how can I git clone the successful tag of source and make install?