Signs that this is all about pr rep talk, considering true;
-Nvidia is suddenly playing nice with customers, extending compatibility of previous generation gpus to future features. Doubtful.
-Nvidia is not gaining any advantage from this efficiency trend while the most of the enhancement is for the lower spectrum of cpus, contrary to its brand positioning. Defaults on current knowledge already.
-If successful, AMD would actually benefit more: Atomic operations allow AMD to inherit multiarray approach that is faster on AMD hardware. Gtx 680 cannot use this path, the fastest is "persistent mapped". Notice, Nvidia has stated that the uav functions that defaulted their 11.1 support status do not relate to their 3D performance in any way. I find that hilarious.GDC 2014 - OpenGL Conference
Just take a look at the slides 34 and 66 please. Clearly, Nvidia does not support building data structures on the gpu
as the "fastest" approach, as persistent mapped relies on the cpu to do the job. Ergo, I safely claim that Nvidia's achievement will never be as efficient as AMD's. Haha, Nvidia.
I have got you this time.Edited by mtcn77 - 3/23/14 at 1:49pm