Hardware requirements:

Intel/AMD 64 bit architecture aka x86_64 architecture with AVX2 instructions.

Software requirements:

  1. ONEDNN https://github.com/oneapi-src/oneDNN
  2. Windows ( Windows 10, 11 supported)

    On Windows, run command %echo PATH% to get a list of directories where os would look for .dll files.

    Linux Glibc >= 2.27, ldd --version to know your glibc .

    1. Download from https://github.com/oneapi-src/oneDNN/releases/download/v2.1/dnnl_lnx_2.1.0_cpu_gomp.tgz
    2. Copy shared libraries following format libdnnl.so.x.x to system's Shared Library PATH.

    On Linux, downloaded shared .so files have to be copied into PATH like /usr/lib or /usr/local/lib for linker/loader to be able to find it.

  3. OPENBLAS https://github.com/xianyi/OpenBLAS/
  4. Windows ( Windows 10, 11 supported)

    Linux Glibc > 2.27, ldd --version to know your glibc .

    1. Install it using package manager.
    2. On ubuntu it should be enough sudo apt-get install libopenblas-dev


Hardware requirements:

Nvidia/Cuda graphic cards are supported for now.

Software requirements:

  • Windows 10,11

  • Linux (Glibc >= 2.27)
  • CUDA TOOLKIT >= 10.2
  • CUDNN Tested with CUDNN_7.x.x, CUDNN_8.x.x
  • Download Corresponding CUDNN version based on the installed CUDA-TOOLKIT version and OPERATING-SYSTEM.

    Make sure downloaded CUDNN is compatible with CUDA-TOOLKIT installed.

Hybrid (Intel/AMD + GPU(CUDA/NVIDIA)):

All the requirements for both of the compute platforms as mentioned above must be fulfilled for hybrid models.

If you are experiencing issues during installation of dependencies please reach out to support@ramanlabs.in .