Saw this http://forums.adobe.com/message/5487822
which has a copy-paste reply from http://www.dvinfo.net/forum/adobe-creative-suite/517364-what-do-i-need-add-adobe-maximus-configuration-2.html
There are a number of misleading aspects in the chart they publish on their website, that makes you wonder how much of the truth they want to hide and how realistic these results are.
Apart from the fact that the system they used for coming up with these results is severely dated and underpowered (a single Xeon E5530 with only 12 GB memory and a single Velociraptor make a terrible combination, that would rank about # 1100 - # 1200 out of 1300 systems in our benchmark), they compared using a highly flawed build #294 of Premiere Pro with a high emphasis on Lumetri effects, that still causes lots of problems with the released version of PR.
So one may wonder if the basis for this published comparison, system and software, is a trustworthy one.
The second problem I have with this comparison is the FirePro W5000 versus the Quadro K2000.
The W5000 has a memory bandwidth of 103 GB/s, the Quadro K2000 only 64 GB/s, so even if one could directly compare Open CL acceleration to CUDA acceleration, it would be a rather useless comparison, because the K2000 would be around 60% slower to start with.
Disappointing that AMD claims only 40 - 45% more efficiency. It should have been closer to 60% if Open CL were just as efficient as CUDA, but that is not yet the case. Don't forget that CUDA acceleration has been around for several years and nVidia/Adobe have been working on optimizing the code continuously, Open CL is a new kid on the block, still in its infancy.
However, if the comparison had been done on a more powerful system you would have gotten much more reliable results. Doing this test on such a weak test platform means that there is a significant imbalance that holds back video card performance.
What I'm also missing is the comparison to a GTX 770, that is lower priced and has a memory bandwidth of 224.3 GB/s and thus starts out with a performance advantage of around 350% over the K2000 and 220% over the FirePro W5000.
Well, that is what Marketing is all about, obfuscating things and only telling half truths.
How much of that bench is memory bandwidth dependent?
W5000 (not the DVI one) has 256-bit bus , 102GB/s
K2000 has 128-bit bus, 64.0 GB/s
W7000 also has 256-bit bus , ECC VRAM , 154 GB/s
K4000 has 192-bit bus , 135 GB/s
The GTX 660 (which is more or less underclocked for K4000) is faster than HD7850 (cut down for W5000) at stock so the linked review seems pointless.
Edited by AlphaC - 7/11/13 at 11:06am