Tesla M40 compare to 2060, 2080, 4060 for ML

I'm building an inexpensive starter computer to start learning ML and came across cheap Tesla M40\\P40 24Gb RAM graphics cards.

Question: is it worth taking them now or to take something from this to begin with: 2060 12Gb, 2080 8Gb or 40608Gb?

If we compare the speed on the chart, they are 40% - 84% faster than the M40, but I suspect that everything will be different for ML. Who has such an experience?

building cards computer deeplearning graphics graphics cards question something speed tesla them