Hagedoorn, Hilbert, NVIDIA Geforce GTX TITAN Z performance figures leak, 04/30/2014, 3:08 PM.
http://www.guru3d.com/news_story/nvidia_geforce_gtx_titan_z_performance_figures_leak,2.html
Looking at this, 2nd graph at the bottom, graph indicates that GTX Titan-Z does 1.6x more performance (in fps) of a GTX Titan Black. This could also imply that both GPU on the GTX Titan Z don't completely scale 2 to 1 graphic card. It's less.
@ arteries,
64bit floating point precision first derivative, as mention by others, has nothing to do with gaming. The reason for this is because when frames are being rendered, the game isn't telling the GPU that it needs to render a certain, specific shade of color x on the fly at certain sections of a single frame. 64 bit floating point precision, is mainly used for CGI and rendering. There are other uses for DP. Besides ECCM and any form for Geo boosting, DP just allows your workstation to render deeper, rich colors on the fly, for individual frames faster than someone who doesn't have the DP power on their graphic card. For example, if you use 32 bpc, or bits per color on Adobe After Effects, this slows your productivity down on the rendering time because you can't process deeper colors faster on a normal card quicker (especially on an AVI extension). Toss in Raytracing, and 6GB footage of your game play, about 30 mins to an hour depending on what recording software you used, and it's going to take a lot longer to rendering the same video on a GTX 780 or 780 Ti. DP reduces the time it takes to produce frames in this scenario. This is more so if you're using Autodesk Maya or Softimage in their Particle effect and rendering section of their software. A 220 frame CGI rendering (more then 7 seconds worth of frames) of fire effects would take about 9 hours or more hours, to render on Softimage at over 30,000 particles or voxels, on the CPU alone. With DP and other stuff, the work would probably cut down significantly. Looking at roughly less than an 20 minutes.
To me, the justification for the price of a GTX Titan Z would have been meet if it had ECCM and all the trimmings like a normal workstation card. Otherwise, to me, this is nothing more than a 3-slot, paperweight of fails on NVidia's end. I'd rather spend the extra $1,000.00 for the AMD W9100 for CGI Rendering.
It doesn't make any sense does it?
My theory is that NVidia felt, prior to release, that AMD was going to flop on it's rear with the AMD R9-295x like so many NVidia, polarized fan-boys were preaching and hoping. If AMD did fail at it, in theory, AMD would have a crappy Dual GPU graphic card, and the $3,000.00 justification wouldn't seem so bad on NVidia's end. If the competition couldn't deliver a functioning product, consumers would throw a fit, and they would be more willing to purchase the alternative. Look at AMD's track record with the AMD 7990 graphic card. Ya it was still the King of the Dual GPU race, but still, it came with a boat load of frame dropping issues. AMD learned their lesson. Not only did they fix their Frame Time issues with a software fix, they provided a Dual GPU card with full Hawaii Cores (actually 90% like the R9-290x) close to it's 1.0 GHz turbo at a slightly higher TDP. Since AMD didn't drop the ball, NVidia couldn't bank big revenue returns for their $3,000.00 new wonder.
@ anyone,
So in truth, your best bet is to have 2 GTX Titan Blacks or 2 GTX 780 Ti's in SLI for gaming mainly. Nothing major has changed with the GPU on the Titan Z. Titan Z is just two Titan Blacks on one PCB with lower Core boost clocks. If 2 GTX Titan Blacks in SLI have a higher Boost clock then a GTX Titan-Z, more than likely, the SLI pair will outperform the dual GPU per card setup. GTX Titan Black, GTX 780 Ti, and GTX Titan-Z all have the same 2880 Cuda Core Count. The only difference is their speed and frame buffer size. There's really no huge performance gains between the three.
