I am interested in working with the Intel Phi, but I wasn't sure if there have been any significant changes in how they are used in the past couple years. I see plenty of them on eBay for around $500, but something I read last week made me think that while the older versions HAD to have the OS loaded on them, newer versions may work similar to adding more memory to your computer, where it would simply show up as being available to the OS.
I have been working with video encoding using an Nvidia GPU, and the available information made it appear that Intel Phi would not be a good fit for ffmpeg encoding, but the next stage of the project includes speech recognition which I understand would be appropriate to utilize the Phi.
That being said, if a newer card is easier to use, it would be worth the investment, but if there is no difference in how they behave, I would test the project on an older card and only upgrade when necessary. I wasn't sure how relevant videos from 2013 are to todays Intel Phi (https://software.intel.com/en-us/videos/part-1-of-4-introduction-to-the-intel-xeon-phi-coprocessor).
The second-generation Xeon Phi ("Xeon Phi x200", or Knights Landing, or KNL) is currently only available as a standalone compute node -- there is currently no "coprocessor" version of the second-generation Xeon Phi. A listing of currently announced products is at http://www.intel.com/content/www/us/en/processors/xeon/xeon-phi-detail.html