NVIDIA_GPU - kileyhartigan/freemocap GitHub Wiki
Background - A GPU (graphics processing unit) is basically thousands of tiny super wimpy processors, while your CPU is one giant, super powerful processors. A CPU runs most of the computations on the computer, but they need to process everything serially (one thing at a time). This is good for handling relatively small numbers of complicated computations, but bad for processing large numbers of simple calculations (i.e. the geometric calculations needed to update the viewpoint of a video game player as they move through a 3D scene, which basically means you have to perform one very simple calculation for each pixel of your monitor - so many many very simple computations ). GPUs are used for graphics, because they basically allow you to do many many simple computations at once (i.e. 'in parallel')
So basically, gamers and animators have been pushing the development of faster, more powerful GPUs for decades, while driving the prices down (so that they can be bought as consumer products)
And as it turns out, the kinds of calculations you need to do a lot of modern Machine Learning/Neural Networks/AI computations (like the ones that underly systems like OpenPose) fall into the category of "Many Many simple calculations" - and so are very well suited for GPU processing
So you COULD technically run OpenPose on your CPU, but you would need to do each of those simple calculations one at a time, whereas on your GPU you can do them all at once - hence why the GPU version of OpenPose is massively faster than the CPU version
HOWEVER - A downside - If you ask your CPU to do a very difficult problem it will break it up into its constituent sub-calculations and run them one at a time. It might take a very long time to finish, but it will finish
If you ask your GPU to run a process that is too large to fit into it's memory, it will just fail.
NVIDIA is just an injustry standard type of GPU.
So, to make sure my computer had a NVIDIA GPU, I googled it ("do I have a nvidia gpu windows?")...and turns out it had one...albeit an older version, but it'll do.