CUDA GPU problems.

[janns]

Hi, I have a brand new nVidia GTX 780 card and get this error message when trying to do a cycles render in GPU mode: "CUDA binery kernel for this graphics card compute capabillity (3.5)not found." I'm new to Blender and would be greatful for some help.
Requards
Janns

[janns]

Forgot to say that I'ts a windows 8.

[fr@nck]

Hi Janns,

I'm looking for an answer too.

For the time being I have an ATI 7870 without GPU support in rendering ...

So I would like buy the new GTX 780 but before I need to be sure that it will work with Blender.

If someone can confirm that GTX 780 works fine (GPU rendering) with Blender it will be really appreciated

Thank you in advance

Fr@nck

[janns]

Hi, I found this revision 57101 http://graphicall.org/444 and now it works great!

[fr@nck]

Gooooooooooood newss

Thank you Janns

[TheArfMan]

only works for the 64bit windows version. I still can't find a build supporting the GTX 780 on Windows 32 bit. Anyone?

[bluemonkmn]

I can't read the error message I'm seeing very well because it only appears for a fraction of a second each time I click Render (where can I see the full message more easily?) but when I try to use CUDA with Device = GPU Compute and Feature Set = Experimental, I get no output and the followign message briefly appears:

CUDA device requires compute capability 2.0 or up, found 1.1. Your GPU is not supported.

I could accept this if there were some information about what *is* supported, but I can't find any reliable information out there. Should I be able to use the CUDA processors on my GeForce 9500 GT? I found the setting to enable CUDA and select the device in User Preferences, and I turned on GPU Compute and Experimental in the Camera's Render settings. But it doesn't seem to be willing to even try to generate any output. Ar the requirements documented anywhere?

[bluemonkmn]

Forgot to mention I am using 64-bit Windows 7 and installed Blender 2.67 this morning (new).

[Tehrasha]

CUDA device requires compute capability 2.0 or up, found 1.1. Your GPU is not supported.

I could accept this if there were some information about what *is* supported, but I can't find any reliable information out there.

https://developer.nvidia.com/cuda-gpus

http://en.wikipedia.org/wiki/CUDA

[bluemonkmn]

https://developer.nvidia.com/cuda-gpus

http://en.wikipedia.org/wiki/CUDA

That says what versions of CUDA compute capability are associated with which hardware, but not what Blender supports. I see mixed information about what version of CUDA compute capability Blender can work with.

[Tehrasha]

https://developer.nvidia.com/cuda-gpus

http://en.wikipedia.org/wiki/CUDA

That says what versions of CUDA compute capability are associated with which hardware, but not what Blender supports. I see mixed information about what version of CUDA compute capability Blender can work with.

I thought the error that Blender spit out should have clarified that.

[bluemonkmn]

I thought the error that Blender spit out should have clarified that.

You'd think so, but the fact that the error is barely visible combined with conflicting evidence I see at http://graphicall.org/444 (references to 'in system console is says " Cuda device supported only with compute compatibility 1.3 or up, found 1.1"') and http://www.blenderartist.org/forum/show ... -GFX-cards ("it just says I must have 1,1 or later") makes me want to know exactly what is supported and what I might be able to do to get either get Blender closer to working with 1.1 or get my system closer to supporting 2.0 or whatever Blender *really* needs. Some indications on pages I've found are that the Experimental setting is supposed to get past messages like that, so I'm looking for some history and more specifics on the CUDA compute capability to see if maybe there *is* a way to make it work as some others have in similar circumstances. I'm hoping that a better understanding would also include some indication of whether I will be able to use CUDA GPU rendering on a future version of Blender with my current hardware.

[Tehrasha]

In that case you need to specify.

As you have discovered, there is a difference between "What does Blender support?" and "What does $custom_build_of_Blender support?"
Or even, "Will Blender support $card in the future?"

You have already discovered the answer to question 1.
Question 2 should be covered in the description of whichever custom build.
But if it doesnt work with your particular card, that is something you would have to take up with whoever built that version to get it ironed out.
Question 3 is the tough one and likely depends on how may codemonkeys have access to $card, or how popular $card is to go through the trouble of supporting it.

What is really needed is some sort of chart of diminishing returns.
What CPU level outperforms $old_GPU.

[bluemonkmn]

In that case you need to specify.

As you have discovered, there is a difference between "What does Blender support?" and "What does $custom_build_of_Blender support?"
Or even, "Will Blender support $card in the future?"

You have already discovered the answer to question 1.

I'm not entirely sure I have. Why are others reporting that Blender requires 1.1 or 1.3 or it varies based on the "supported" setting? I'm not sure what the variables are that determine what CUDA is supported under what circumstances.

[farbw3rk]

[quote="bluemonkmn"]I can't read the error message I'm seeing very well because it only appears for a fraction of a second each time I click Render (where can I see the full message more easily?) [/quote]
have you tried rendering by commandline?