Aua, this is new, did you file a bugreport?
The card is not officially supported but it is may good information for Cycles dev´s.
Cheers, mib.
After reading this,
h**ps://developer.blender.org/rB2d26d1dbbd0e91c2fd7b6dd147ab54dbd7f73111
May be a naive question, but…
Can be as simple as this if I compile my own 2.69-32bit build?
(Sorry if I’m clueless)
I’ve mentioned it a few days ago DingTo,
because I think it affects all cards from sm_35.
I could narrow it down a bit and reproduce with a SM20 kernel on the GTX570, which was built with the CUDA toolkit v5.0
but I could not test with official build on sm_35 cards.
But I just noticed, there is now a bug report, which confirms my theory => https://developer.blender.org/T39247
This isn’t new but there is apparently only too few 780ti/Titan owner 
You can compile your own build. But if you’re like me (probably) and don’t have much knowledge about the process, it can be tedious and endlessly time consuming.:spin: What’s worse, if it compiles incorrectly, all that effort goes wasted.
I guess I’ll sit tight till the new 2.70 release is out. Unless rolf can be a very awesome guy and give us a 32-bit OS build:p
Bug is fixed and 2.70 is out now, so here some updates
Blender 2.70 64-bit Windows with sm_50 kernel Hash: cc2cdfb
Blender 2.70 32-bit Windows with sm_50 kernel Hash 19fcb4d
http://goo.gl/ZfD2rY
and for testing a special sm_50 kernel, this kernel enabled subsurface scattering and volume scatter/absorption on 750/ti, only for 64bit.
Vram usage is much higher with this kernel.
Put this file to …Blender\2.70\scripts\addons\cycles\lib
I just bought the gtx750ti and was a little frustrated.
Now, I can jump for joy!
(The viewport in cycles render mode it’s amazing for me!)
Yesterday I changed to x64 one of my computers only for test your previous version 2.69.11 (R61319) and works like a champ also.
That’s great, because now I can test your “subsurface scattering and volume scatter/absorption” that I didn’t expect!
Can we ask for more?
We owe you some beers, friend.
(especially ‘The Sun’ and I)
Thanks again for sharing, really.
( there is a folder with your name ‘Rolf’ in my PC for tribute 
I have also compiled sm_50 kernels for release (both 32 bit and 64 bit)windows http://martijnberger.nl/file/2.70_sm50.zip
linux http://martijnberger.nl/file/2.70_sm50.tar.gz
extract into 2.70\scripts\addons\cycles\lib\ windows
Working perfectly! Thank you, rolf!
Just one small question:
Will the sm_50 kernel builds be distributed separately, like some guys said, or will it be a part of future Blender releases?
anyway, thanks!
Using ROLF’s build
AMD FX 6300 Vishera 3.5ghz
8gb ram
with the EVGA 750 ti FTW
1:18s for Pan’s BMW
When I learn anything about overclocking i’ll give it a go and post it.
Hello.
Here are my times for my GTX 750 ti running in Windows 7, 8Gb Ram Intel i7 3.5 Ghz, and the latest build of Blender http://www.graphicall.org/110
(all using GPU)
BMW1M-MikePan: 1:16:35 (Default settings)
cornell_bench: 2:18:04 (Default Settings)
pabellon_barcelona :too long. 15 minutes at 18/28 Tiles at default 1000 samples I stopped rendering.
Is everyone using the pabellon with its default 1000 samples?
Here is pabellon_barcelona at 100 Samples: 3:08:45
It seems like my times are slightly slower than everyone else. I wonder if that has anything to do with the particular build of blender that I am running.
750ti using Rolf’s build.
(thanks Rolf)
Mikes BMW: 1min 14 seconds.
Cornell room: 2min 08 seconds.
I just got my 750ti today and i am ecstatic !!
So fast, so quiet, and it doesnt even heat up.
Hi 750 user, could you render my new bench with your cards, please?
http://www.blenderartists.org/forum/showthread.php?331149-The-new-Cyles-GPU-2-70-Benchmark
Cheers, mib.
Thanks alot - linux kernel works like a charm 
sm_50 kernels working great. Here’s my times using the 750Ti:
BMW Mike Pan - 01:20:71 (Default settings)
Cornell Bench_27 - 02:30:49 (Default settings)
Thank you,Rolf.
Here’s my benchmark of a 2048MB Palit GeForce GTX 750 Ti StormX OC Dual Aktiv PCIe 3.0 x16 on AMD FX-6300.
Hi Jekyll&Hyde, can you please add your result to the benchmark thread?
http://www.blenderartists.org/forum/showthread.php?331149-The-new-Cyles-GPU-2-70-Benchmark
Thanks, mib
Scotchtapeworm impressive time, by the way on my new 750 ti , and I7 870, 16 Gb RAM:
cornell_bench: 3:00:04 (Default Settings) ,
the same time on 2.69 and 2.70 Rolf’s version. on the latest version it does not work( although I copied sm_ 50 kernel )
Hi, it’s an interesting work you’re doing here.
I tried my GTX 750ti (2GB MSI OC) with i5-4670 and 8GB RAM
Mike Pan’s BMW (default) 01:00.49
Mike Pan’s BMW (256x256) 00:59.46
Cornell Box (default) 02:58.81
I used 2.70a with sm_50 kernel for the testing
My GTX570 died, so i get a another 750ti.
I get back to 2.69.11 for now, the rendertimes are worse on recent builds for me(80s for BMW). And with two GPU´s i get only a speedup around ~33%.
On 2.69.11 i get a speedup near 100%.
I’ve render the Caterpillar Scene on 2.69.11 eb4f2b4 with SSS enable, for only this render i used 3x750ti. yes 3x.
And this is the result:
00h 56m 12s
In 2.69.11 with one GPU i get times around 2h 40m.
With newest 2.70.2 i can’t beat 2h 40m even if i use 2 GPU’s