Take it with a grain of salt, but that would be one crazy card indeed. I'm uncertain how much crossfire support to expect from developers moving forward, as both SLI and Crossfire seem to be falling by the wayside.
Good points on the more than 2xGPU Crossfire support. I was hoping to see a combined GPU article, but if it's Crossfire based it probably may not be as popular, sadly. The only other issue I have with this article is the line:
"The liquid-cooled RX Vega 64 has a TDP of 350W, so a dual GPU card would likely draw at least 600W."
TDP does not necessarily equal power draw, especially when comparing between AMD and Nvidia or even between AIB partners. AMD and Intel (or Nvidia in this case) use different ways to display TDP so a direct comparison between the two may not always be best thing to do - if for some reason you even care about TDP. I know this is not related to your post, but wanted to mention this as a lot of people over emphasize TDP comparisons without actually understanding the variations.
With that said, I am curious to see actual numbers at the wall when I get my RX Vega's up and running; if it's more than I'm comfortable with and the performance benefit is low to non-existent over a single card set up, I'll just build another system. So far I'm pleased with the overall performance with my RX 480 Crossfire setup. Everything I run on a daily basis sees a clear performance boost over a single card; RX Vega may very well change the need for that.
Indeed, TDP usually refers to the level of energy lost by a CPU/GPU as heat as opposed to doing electrical work. The wattage pulled from an outlet will always be a combination of both the wattage lost due to energy transfer, and those used to do the actual electrical work of the system. The energy transfer wattage is important to measure, as it will dictate the type of cooling solution necessary to maintain operating efficiency of the silicon in question.
You are correct that a substantial number of online publications incorrectly treat TDP wattage as the power draw from a metered electrical system.
As far as crossfire goes, I haven't really found it to be necessary. As I haven't stepped up to a 4K monitor, my single Fury X has been performing just fine, as did the HD7970 and HD5870 that preceded them. I am altering my setup a bit currently, as "Heroes of the Storm" and a few other games are just easier to play from a desk.
I do think that Vega will finally be an AMD product that will really benefit from the amazing custom loop cooling you have going on. Fiji and Ryzen both hit architectural limits to overclocking long before heat was a problem. Vega however seems to be able to push as high as your cooler can handle.
After mentioning Crossfire, and looking at Threadripper reviews I am starting to wonder if a double GPU Vega card will use on board crossfire or if it will stick two Vega dies together on infinity fabric the way Ryzen CPUs do. I am unaware of the extent of infinity fabric implementation in Vega, but I have read that Navi will likely use that sort of implementation. A small core GPU that you can glue together on infinity fabric for larger and larger GPUs, killing the need for developer support to see scaling as in Crossfire.
Really interested to hear more about this card. If it in fact does use infinity fabric to join two Vega GPUs instead of two discrete dies joined by an on board crossfire link, the scaling would likely be near 100% from a single GPU in most applications.