win 8 64bit
MSI gtx970 @ stock clock
256 x 256
9:07.46
Here comes an oldie
Blender: 2.72.2 2491dc7
CUDA Version: 6.0
CUDA Arch: sm_20
Card: GTX 480 SC 1.5GB @ Stock Clock
OS: openSUSE 12.3 64bit
Tile: 360x810
Time: 15:24.34
Edit: woa my 1K post, it only took me 10 years 
OS : Win 8.1 64 bits
GPU : GTX 780M, 4Gb
Render time : 14:45.10
Just for comparison from an older pc
OS: Win7-64
GPU: GTX460/2gb
Time: 23m53s
os: Win7-64
gpu1: zotac GTX 680 4gb
gpu2: zotac GTX 780 Ti 3gb
time: 05:38.93
os: Win7-64
gpu1: nvidia GTX 590 3gb
time: 07:57.31
I’m sorry but I think the image is not very “sexy” , yes I know it’s a benchmark, but I rather tweak Mike Pan’s BMW and modernize it to modern GPU render times.
Maybe for V3
if I am not mistaken the Titans are still rocking all other cards. I was quite stunned to see that the 970 which I am planing to buy is nt that much faster than the 570 I have right now.
Mac OS X 10.10.1
GPU - EVGA OC+ GTX 660 3GB
Time - 16:59:33
Im surprised my graphics card is that crap in comparison to the others on the list >.>
Crunchbang Linux
gpu1: - GTX 580
gpu2: - GTX 970
Time 4:59.65
My Results:
OS: Windows 7 Professional 64 bit
GTX 680 4 GB
GTX 680 4 GB
Both cards
05:49.25
I see you are using the 750 ti with Mint 17. I plan on getting a 750 ti as well at some time and I have Linux Mint 17.1. How do you like using the 750 Ti with Blender and are there any problems you have had with it?
Windows 7 x64
GTX 590
7:15
Got myself a secondhand 590 for £75. Nearly three times faster than my 650 Ti Boost on this benchmark. That’s pretty good value for money, I reckon.
My results:
1x Gigabyte GTX Titan Z 12GB DDR5 (watercooled without overclock)
03:09.76
Edit: Windows 8.1 x64, Blender 2.72
Windows7 x64
GeForce GTX 760 4GB GDDR5
13:16.21
Blender 2.72
Blender 2.72b
OS: Linux Mint 17.1 - 64 bit
Card: GEForce GTX 670
Driver: 331.113
Time: 16:09:39
I really like this benchmark list, but (as mentioned earlier) the choice for 4 tiles has some serious consequences for the top-rating cards: It now seems to suggest that it is better to have two superclocked 780Ti’s than three 780s or Titans, while actually the use of 4 tiles is causing two of the three graphics cards to do nothing half the time.
On the other hand, changing the benchmark would invalidate all the results people have committed already.
Could we have a re-benchmarking for people that have multiple graphics cards? (using e.g. 12 tiles). That would also allow better load distribution for people that have a system with one slow and one fast gpu.
Blender 2.72
W7X64
AMD HD7970
Catalyst 14.4
09:16:39
What ? AMD works perfectly ?
yes, but only in GPU mode… gpu+cpu -> blender crashes
EDIT - no it’s not perfect - sky texture missing, volumetric fog missing… 