It is certainly more complex, but hopefully Apple can get AMD to help them with the drivers.
Nvidia and AND already do the low level drivers. There is little to no evidence that "Crossfire/SLI was too complicated for Apple to figure out"
so Apple didn't do it. Far more likely it was ( and still isn't ) done because getting the far more broadbase drivers done and high quality is a much higher priority.
It just seems silly to have hardware (rather expensive hardware) in the computer that many people who buy the computer won't use.
Why is it not going to be used? You presenting the case that as API abstraction of a virtual device be put in so that applications can transparently leverage it and make use of dual cards. Just as easily an API ( GCD , OpenCL ) can be put use by the corer library and apps. Apple has move core graphics calls either up into CPU or down into GPU on previous OS X updates. The 10.9 updates to the OpenCL drivers have hooks for heterogenous memory access across PCI-e so data can be pulled in and computed without explicit copies (i.e., Apple has pushed agressively for some pre OpenCL 2.0 aspects early )
Also talking about non use over most of the service life of the machine.
I think that if Apple weren't looking to implement SLI/CF in the near future, they would be giving the option of a single card, as opposed to dual cards standard.
Configurations like this:
http://www.nvidia.com/object/maximus.html
have been becoming more common in the Workstation market over the last couple of years. It isn't a "just started" trendline.
The other trend line coming is that once again monitors will drift back toward a 1:1 ratio with video cards. Monitor resolution being stagnant while GPUs get increasingly faster is over.
Maybe it won't be there on release, but I think it will, as 'one more thing' Mavericks has.
Highly unlikely that Mavericks has "one more thing" that isn't some graphics visual veneer thing. Mavericks will get some upgrades downstream 10.9.4 or 10.9.5 that has upates driven by new hardware, but doubtful there are major features being suppressed. On a yearly upgrade cycle there is little to no reason to do that. Major updates are constantly rolling out in next 12 months anyway. Extremely better to put the full OS out. Let developers/beta folks gets the bugs out. Release. Get more bugs out. Then go no to next update. rise and repeat.
The main reason that up until now I thought there would be no CF is that there was no bridge visible in the nMP pics. Now with AMD saying they are no longer required, I think it will happen.
The connectors on the whole card are different from the rest of the discrete card market. There is no evidence of socketed edge connectors. The PCIe connect doesn't look to be on the "long" edge. Looking for common mainstream connectors when it isn't a mainstream design isn't going to say much one way or the other.
If Apple got to market where they are selling 500K of these a year and 100+K folks of that set need super high frame rates ... it could possiblly happen. Far more likely is that the numbers will be much lower than that. Ultra high gamer frame rates aren't going to sell many more systems for the additional complexity.
That's not really a fair comparison. The old MPs came with options for multiple low powered GPUs, purely (and obviously) so they could use more than one monitor.
Apple hyped 4K. Multiple 4K monitors means????? Even one 4K monitor will be better served by moving non graphics required for just that specific monitor to somewhere else (e.g., secondary screens, computations, etc. )
I do not think that Apple have ever provided the option of having two performance-enthusiast level GPUs.
This isn't aimed at "enthusiast". It is aimed at folks who have computational work to do.
The comparison with the iMac and 15" MBPs are obviously flawed, because they have different GPUs (which would be impossible to CF/SLI) purely for graphics switching, performance vs power saving reasons.
There is a realm of usefulness where both cards are not locked into doing exactly the same thing. In that dimension it is the largely the same. Two different uses for the two GPUs. There is no reason other than the "two entry" and "two top level" card configurations that the two GPUs cards in the Mac Pro have to be the same two FirePros. In fact, it be surprised that outside the lowest/highest configurations that they are.
Now, as my final note, I am hoping that when the nMP is released, it at least has an option for two top end Hawaii cards.
That would be nice. It just doesn't match up with the highly consistent track record for both "Pro" card releases and "OS X specific driver" releases.
Far closer to the track record in 2010 when Mac Pro shipped ( July) with the HD 5x70 series when the 68xx series shipped 2-3 months ( October).
A slightly more realistic hope is that the GPUs (and CPU) will be replaceable,
Extremely likely. That is primarily directed at repair/serviceability. However, ....
and there is an indication that either Apple, or 3rd parties will be creating and selling (with a hopefully not too unreasonable markup of course) new Apple proprietary GPUs. So we can later buy 2x R9 290X's and slap 'em in.
Later ( after 2014 Mac Pro are old and start to be retired and sold for parts) yes. Short term? Probably not. Especially if folks "rant" complain to Apple for a solution.
Apple doesn't build "alternative" GPUs for iMac , MBP . They'll switch suppliers every generation or two if something better shows up. Same issue here. It is extremely doubtful they are just going to willy nilly approve stuff that isn't highly designed to be coupled to their thermal solution to be slapped into the new basline Mac Pro design. 3rd party daughtercards never came for the 2009-2012 Mac Pros. These GPU daughter cards are in the same general class. This new design is primarily built about there
not being arbitrary components being slapped inside the box.