Problem rendering with a GTX 590

[PhilEdgar45]

Hi,
I have just recently installed a GTX590 for faster rendering with Octane and have come up against a problem straight away. I am using the correct version of Octane and the correct drivers so now need a bit of help of sorting it out.
When i render a scene i get the rendered image split in two and one side is clearing faster than the other - it is as though the image is being rendered by two different cards of different power. Also the amount of CUDA core available in my preferences is 512 - is this correct?

Any help would be appreciated as deadlines are fast approaching...
Many thanks.

[Cronicash]

I still believe using 3.0 (only multi gpu version) you wont be using all cores. I notice this too, if you have a scene where one part is more complex than the other say more bounces required for a clear picture due to emission or translucent it tends to hinder the card dedicated to render that side. Not sure if this is what developers intended but that as far as i can figure...? Is this right? I get the same with four cards, just occurs in quaters...

Is there a way to make the cards render in an interlaced method for future updates? (probably would be less beneficial to do so for tedious renders)

[abstrax]

We are currently on a complete rewrite of the multi-GPU stuff, which will solve these problems.

By the way, could you please post an image that shows this problem, because I can't reproduce these effects at work? I also would like to know which build exactly you are using and which driver version.

Cheers,
Marcus

[PhilEdgar45]

Hi,
i am using the version 1.0 beta 2.46b with the Nvidia driver 270.61 which is CUDA version 4.0 (I only recently updated the nvidia driver and i was getting the same problem with CUDA 3.2)
i have attached a screen grab of the problem.

1. is this an Octane problem or a problem with my settings?
2. i have tried multi gpu disabled and get the same problem. Does this mean anything?
3. The speed is about the same with or without multi gpu. Is this correct?
4. if this is an Octane problem and the new rewrite will sort it, is there a workaround to get a single card speed without the split image while we wait?

Thanks

[PhilEdgar45]

Forgot to mention i am using the 3.0 build for multi gpu use...
Sorry.

[abstrax]

Ah yes. I have seen that before: For some reason the OpenGL texture (/pixel) sampling works differently for different GPUs when the zoom factor is <100%. If you zoom the viewport to 100% or higher the noise should be exactly the same. We haven't really investigated this problem yet, because it's only a display issue. If you save the image the noise should be equal, too.

Cheers,
Marcus

[PhilEdgar45]

OMG you're right.. That's saved my a*se!
Thanks.

[Timmaigh]

Just wanted to start a new topic about this, just yesterday i was rendering with my GTX590 and ran into exactly same issue with the 3840x2400 resolution. Good to hear, its just a "cosmetic problem, not some serious performance drawback.
Anyway it brought another thing to my attention, i had a big part of the screen really really clean with only tiny noise, however there was a "dark" corner, which obviously required more time, and because of this corner i had to render maybe 2x longer... iit would be perfect, if you could somehow "concentrate" your rendering power into regions of screen you wish, i saw this in the Iray promotional video, so i suppose its possible to do (as they claim basically they do same thing as Octane, with the -maybe crucial- exception), they are hybrid renderer.

[kivig]

iit would be perfect, if you could somehow "concentrate" your rendering power into regions of screen you wish, i saw this in the Iray promotional video, so i suppose its possible to do (as they claim basically they do same thing as Octane, with the -maybe crucial- exception), they are hybrid renderer.

+1

IMHO a feathered vector mask and lo/hi proportion parameter would do great.
If any sort of animation will come into Octane, it'd be awesome to animate those mask paths like in AfterEffects.