CUDA and vector displacement

Hi,

not sure if its a bug or lack of support but today for the first time I tried out CUDA and the scene happens to use vector displacement.

For some reason it doesn’t render - its basically rendering only the low rez polygon mesh. Is that lack of functionality?

Am I expecting too much from the CUDA implementation?

A bit of a negative first experience with CUDA right off the start I say! hehe :stuck_out_tongue:

thanks
Peter

Can you post the scene or a version of it that will reproduce the problem?
It would also help to know your VRay/Maya/OS versions + nvidia hardware and driver versions.

My specs:
OpenCL 1.2 CUDA 7.5.15.
Driver version: 10.18.13.5382
Quadro K4200 & Geforce GTX 660Ti
Vray Nightly build 26.5.2016
Win10 Pro
I will prepare the scene later the day. But does CUDA vray support vector displacement?

It should work fine, I think - at least I don’t see an immediate reason why it won’t.

Best regards,
Vlado

So i tried again with fresh scene and I come to same conclusion:
Check out the picture with comparison I made.
Oh, btw. I am using 26.5.2016 release of vray

Can you check if the “resize textures” is enabled for GPU (it is by default) and disable it to see if that helps.

Ok. Here is the scene.

Dropbox - Error - Simplify your life

I tried all resolutions and bit depths with and without checkbox on ‘GPU resize textures’

here it is… :slight_smile:

thank you . found it 1 sec after posting :slight_smile: hehe

Well, then all that’s left is to debug the scene. Thanks for the link.

Hello Pete3d,

Vector displacement is a supported feature on the GPU, however the GPU handles displacement a bit differently than the CPU.
At the moment the Bump map that is applied to the mesh after displacement that you can see when rendering on the CPU is not a supported feature on the GPU.
Also, the GPU requires normals to be calculated before rendering and cached (this is the Cache Expanded option so V-Ray will always consider it ON when rendering on the GPU).
This option relies on the max number of subdivisions to produce good results. You can see how the CPU and GPU results are quite similar if the Cache Expanded option is ticked.
So I will advise to crank the max number of subdivisions (which by default is set to 4) and see how this works for you.

Unfortunately I don’t think I can give you a better solution at the moment.

We will update the help pages to include this information as soon as possible.