I've also found it's difficult to get a clear view of mining performance from the various GPUs. People will show results from heavily overclocked cards, results that are not long-term stable in my experience. This is stupid. If you modify your VBIOS to use tighter latencies on the GDDR5, it can improve performance by 5-10 percent ... and a few weeks later your cards will all start failing. I've been there with the RX 480, and all that downtime just isn't worth the hassle!
My advice is simple: buy a factory overclocked card with a massive cooler, use PCIe risers to separate the cards from each other, and then just run at the default clocks. The cards will run stable, at reasonable temperatures, and you're on your way to profitability. In fact, rather than overclocking, sometimes it's better to underclock and undervolt your card, to reduce power and temperatures. You're far less likely to burn out a fan, or overheat, or any number of other potential problems.
Okay, let's get on to the RX 580 8GB. I'm using the Asus ROS RX 480 Gaming OC, and absolutely massive card that's bigger than many GTX 1080 models. The benefit is it has three large fans, the cooler keeps things frosty, and it comes with a healthy factory overclock. It uses a single 8-pin power connector, and in practice if you install it in a case, you'll use three slots -- you really can't put anything into the two adjacent slots due to the size of the cooler.
And here's how the card performs, at the stock OC, on the various current algorithms:
This is with one of the most expensive RX 580 cards, of course, and you could drop to a 4GB model without really losing any performance. As an alternative to the Asus, the MSI RX 580 Gaming X 4G costs just $215, and it has similar clockspeeds in the default 'Gaming' mode. ROI on a card like this is only 100-110 days, conservatively -- and if Bitcoin prices improve, you might hit ROI in as few as 60 days. Or things could go the other way and it might take 200 days.
Overall, the RX 580 is one of the best options for cryptocurrency mining right now. With a single 8-pin connector, it makes it relatively easy to put four or even six GPUs into a single mining rig. I recommend getting a great power supply for such a configuration, with the Enermax Platimax 1350W sitting at the top of my list.