Does that mean you can't use DP thru OpenCL?
generally no (i.e., yes, it is usually available).
"OpenCL—Code Generation (emphasis added below):
....
Double as single. If you set this parameter to Yes, the compiler treats double-precision floating-point expressions as single-precision floating-point expressions. This option is available for GPUs only.
The default is No. ..."
https://developer.apple.com/library...rogGuide/XCodeHelloWorld/XCodeHelloWorld.html
If they didn't usually support double this flag likely would be default to the other value.
Of the AMDs GPUs that Apple has officially released as part of standard configurations, most have not had DP support (I think only the 5870 ). However, Apple's default is make a distinction, so the infrastructure is there when devices match that presumption.
Even some of the aftermarket cards later 6750 don't.
http://stackoverflow.com/questions/17875843/using-square-root-function-sqrt-with-doubles-in-opencl
Ballpark for the D300 , D500, D700 is likely around 6% , 25% , 25% respectively of the single numbers that Apple quotes ( working backwards from Pitcarin XT , Tahiti LE , and Tahiti XT here
http://en.wikipedia.org/wiki/Compar...g_units#Southern_Islands_.28HD_7xxx.29_Series )
If have serious long computational time DP work to do, then probably don't want the D300. Substantially slower DP perfomance and highly likely no ECC either.
If not, is that likely to change when the new MacPro is released?
Probably yes. First time the entire current AMD GPU line up is all DP capable. They could snore on that, but the blowback from shops that do "accurate physics simulations" (a feature highlighted on Apple's Mac Pro marketing webpage) is going to be pretty intense.