Are Tesla K80s any good for inference?

A Reddit user is asking if Tesla K80 graphics cards are suitable for inference tasks, noting that these cards are available on eBay for $50-60. They are curious if anyone has tried using these GPUs for such purposes, considering their low price makes them an interesting option for experimenting on an extra PCIe port, even if groundbreaking performance is not expected.