Project Goals


This project aims to provide cross platform OpenCL tools for deep learning and inference.

Today, most of deep learning training is done on NVidia GPUs using closed source CUDA and CUDNN libraries. It is either challenging or virtually impossible to use AMD or Intel GPUs. For example: AMD provides ROCm platform, but there is no support of RDNA platforms yet (more than a year since a release), there is no support of APUs and no support of any operating systems other than Linux.


  • Create an open source, cross platform deep learning primitives library similar to cuDNN or MIOpen that supports multiple GPU architectures.
  • Create an inference library with minimal dependencies for efficient inference on any modern GPU, similar to TensorRT or MIGraphX.
  • Create minimalistic deep-learning framework as POC of capabilities and performance.
  • Long Shot: Integrate to existing large scale deep learing projects like PyTorch, TF, MXNet such that vendor independent open-source OpenCL API will be first class citizen for deep learning.

Please note this is only work in progress - first and preliminary stages.