This was the first week I started to work on the second half of my project, the GPU accelerated deep learning on HPCC Systems Platform. HPCC has the capability of using embeded languages in and alongside ECL code. I will use python via the pyembed plugin. Python is a very popular language for many things, one of which is scientific computing such as deep learning. There are many different libraries and packages that make it the language for this purpose. Two of which are TensorFlow and PyTorch.
This week I started with creating a proof of concept single node system. The challenge it to get HPCC systems to play nicely with CUDA (for GPU), and all the various python plugins. Careful installation of the libraries is needed. If HPCC is installed after, the HPCC user does not by default see all the installed packages. So I figured to install HPCC first, then install all python requirements under the HPCC user using $sudo -u hpcc apt-get ....
Comments