How are the nMP GPUs multiplexed?

[Cubemmal]

This was somewhat discussed a while ago, but how do they allocate 2 GPUSs for three ports? According to Apple the bottom two are Controller 0, the two left above those are Controller 1, and the two right are 2.

Say I connect 3 4k monitors, does one GPU get two monitors? Does this imply internal Crossfire usage?

 

[Gav Mack]

PCIe 3.0 supports cards connected via the bus eliminating the need for crossfire. On efficiency alone I'm taking an educated guess they have gone down that road..

 

[VoR]

Not that I've looked closely, but It's a proprietary card and they could build it into the connector, but crossfire isn't required for multi-monitor or ggpu computing.
The 7970 and 280x (rebranded 7970 with GCN) both require a physical bridge, the D700 is just an under-clocked/voltage version of these tahiti based boards.
The hawaii based 290 cards do their crossfire through pci-e (and pci-e 3.0 isn't a requirement for this, 2.0 works just fine).

 

[Cubemmal]

Sure they could run two monitors off one card but that's an asymmetry and difference in performance potentially. I would think Apple would advise us as to which ports got the dedicated GPU.

 

[goMac]

There is no Crossfire support under OS X at present. Each display is driven by a single card.

You can use one card for OpenGL and another for OpenCL within the same app, however.

Source: A little biddy.

I don't know what port goes with what card.