Tested with CUDA too. It’s heavily slower on GPU.
For CPU I used (f12 render) 64x64 tiles size
For GPU 256x256. Nvidia 570
CPU 1 min 40 sec
GPU 3 min 37 sec
Ah, our dear Murphy’s law continues to work. See storm_st?
this is my first time EVER compileing CUDA so…
I would really like someone else to recompile it on VC with CUDA or even
MinGW64 again if possible… maybe i did something wrong…?
like i said… its my first time try to build with CUDA’s. 
i also wondering if 256x256 may not be the best tile size for hair?
have you tested any others?
Wow!! This is BIG!!
Seeing very promising results already.
YES YES YES YES YES!!! AWESOME! I’m so excited for this feature in cycles. Man, those renders look amazing. Great job - keep up the awesome development!
I love this character. Very unusual!!
I would love a linux build 
Bug Report Incoming!
http://www.pasteall.org/blend/17993
Getting this strange circular, almost moire-like artifacting in my render.
It does something strange when I change the width multiplier where it rebuilds the BVH, renders 3 or 4 samples correctly, then automatically starts rebuilding the BVH again for no reason, and then starts rendering this weird circle pattern.
EDIT: On further testing it would seem to be related to my choice of focal length. Setting it from 60 back to 30 eliminates the problem. Although the BVH is still building twice per render.
I’m happier than a kitten following a leaky cow… this thing works great and … FAST. Thank you SOOOO much sir. I’ll start remaking my characters and post as I have a chance.
Not yet since i’m getting quite fast results with CPU. Will try more soon.
I will also add that rending starts with my 560Ti, but is PAINFULLY slow compared to my CPU. I’m not too concerned though, as I wouldn’t be able to fit many hairs into vRAM anyway.
Can’t tell you how happy I am that this is finally in the works!.
I get the same problem on a 560TI - with or without checking the cycles strand rendering. Top of the render window and in the terminal window it shows:
CUDA error: Invalid value in cuMemAlloc(&device_pointer, size)
For the most part it renders fine on GPU when there are strands in the scene, although I did at one point get one of those unresponsive video card errors and Blender quits.
I did also come across a really weird thing that I haven’t yet been able to re-create. At some point when I was playing with the nodes in the material, the 3d viewport flipped upside down and the mouse scroll wheel would only zoom out. All navigation was upside down too. Meaning that the grid floor and all the objects would pivot in the opposite direction the mouse moved in.
wow, this is spectacular!
any chance of just a scrap of a tutorial for the uninitiated??
I copied the node setups I see and enabled Strand render but no joy
drobbins at the bottom of the particle setting there is a setting called
“Cycles Strand Rendering”
Mode (Tangent Shading) and several others
Strand Shape .2
Width Multiplier .4 or .5
try that.
WOW!!! I hope this makes it into trunk someday soon. Thanks for your hard work.
Mac people! Bashi has a build up here! http://graphicall.org/1022 (seems to be 32bit though?)
Found a couple “silly corner case bugs” to be aware of. 1: If you hit render while in particle mode, you get 0 hairs, so switch back to object mode first. And 2: if you set display count (under the display tab, not the children viewport count) to something other than 100%, it crashes Blender.
Other than that, awesome work! Renders quickly, shows in reflections and everything! (some of us have to use mental ray sometimes…)
Getting the circular alien pattern bug too. But is only in f12 rendering.
Viewport screenshoot:
F12 rendering (less samples, nevermind the noise difference)
First the caveat… This “hair” build is the first I’ve used with the new tile rendering options and I have to admit I’m not entirely up to speed on how best to use it.
Had to add the single hair plane, but under the 2.64 regular build I have (number of tiles not size) it did the job in 1:31 min.
The hair build did one 256x256 tile in 1:48, one 640x360 tile (just divided the image size by 3) in 3:43 and if I go much bigger in tile size I get the driver not responding error.
@ Storrboy how about 256x256
(btw, this is for the MikePan scene?)
leave the image size default… 50% and whatever it is… (1920x1080 i think)
just want to know if its even compile right to begin with. 
you can get trunk mingw64 buildbot builds from here(with CUDA)
just to test.
http://builder.blender.org/download/
to install as portable just create a “config” folder inside the 2.64 folder.
unzip it and name it like “Blender 52XXX Mingw buildbot” for instance
doing that you can have different keymaps for different builds and things.
just keeps things from getting funky. if you like to experiment with several builds.
thanks for the reply
I just don’t see what you’re talking about
my guess is something is wrong in my application of the patch and what I’ve built doesn’t have this feature
experimenting, this looks fun to play with
there will be a build on graphicall tomorrow:D