Gtx 1060 or 780 for GPT2 Text generation

I’m trying to decide which would be better for text generation using GPT2, The 1060 has a higher clock speed but the 780 has more cuda cores, which should I go with? Or should I look at something else around $100?

These cards do not possess the memory capacity to fine-tune GPT-2. For other models that require less, the 6GB version of the RTX 1060 would be an affordable card to practice with. Tim Dettmers at least has a bench mark for it:

If this is out of your budget, you likely will be better off settling for free cloud computing services like Google Colab.