GPT is not a public model. You can't train or run it yourself.

I just checked and saw OpenAI does have a fine tuning API so you can fine tune and use the model through their API, but your hardware doesn't matter.

You can look at open source reproductions of GPT like OPT, but it will be very expensive to get the hardware to run the model, let alone train it. If you really want to use one of these huge models yourself (and not through an API) I'd advise starting with AWS before you consider buying any hardware.