Honestly, neither. It gets too complicated when you go down to the pure hardware level. Its better if you look at individual benchmarks of each card and look at how it compares to other cards in the games that you play or plan to play. If a card beats another card 12 out of 20 games but all 12 games are not games you play, then you might as well go with the card that wins the other 8 games that you actually play, because its more relevant to you.
But i realize i may have read your question a bit wrong, if you are referring to cards of the same labeling, like comparing r9 270X's; then its usually a combination of the memory clock and the core clock. but i generally find the core clock speed as a better frame of reference unless the memory clock is significantly higher. that being said same card product from different manufacturers may not perform the same on a clock for clock basis.
I usually browse benchmarks, which can be time consuming :9