OctaneRender™ Standalone 1.14

[pixelrush]

Probably the vram reporting and texture message are misleading. I think I have run into this issue before when I tried to stuff as many polys as I could into my gtx460/2gb for fun. After some experimentation I managed just short of 10 million with one simple mat no textures and a pretty small render resolution. That was a while ago though. I presume the space required is much the same for recent Octane releases. I concluded cuda needs some free space to operate. I think I decided about 7.5% was required. You might be just over the limit with 27 million polys. Maybe you are only 10,000 over budget? Not sure. Try culling some polys here and there in the tires as was suggested (or perhaps missing one out since you can often only see 3 at a time) or setting a really small render resolution to see if that frees up enough mb to get it in. Of course if you do the image size might then be the problem. If you have textures to load as well and intend to render at a large resolution you might be out of luck with this model even with 6 gb vram.
HTH

[abstrax]

Try using instances to lower your poly count.

Tires are the easiest one, you can cut the count down a lot.
Also, look for things in the autocad that are not visible in the view, like under the car, and remove those for the render.

Thank for the support.
I know that but as I mentioned there are tons of dense meshes that I'll add to the scene.
I'm afraid to hit a wall at some point where I cannot reduce more of those meshes.
As it is a commercial project I cannot risk it. If Vray and iray can render it octane should render it, too. Right?

Wrong. Each render engine uses different ways to represent geometry in VRAM. For speed reasons we use textures not only for image textures but also for other things and they have currently a maximum size of 2GB (per texture). There is a way around the problem for compute model 3.5 GPUs (like GTX TITAN), but it's not implemented yet and requires some ground work.

That's why I want to learn that from the developers. If there is a hardcoded limit in Octane core. I'ld like to learn it. And plan my projects according to that.

The limit you are hitting here is very likely the 2GB per-texture limit. Until 1-2 months ago, the largest amount of VRAM you could get for a GTX GPU was 4GB and the 2GB limit wasn't an issue since you would hit the total memory limit before you would hit the per-texture limit. -> If you don't want to reduce the amount of geometry then you probably have to resort to another render engine that can handle the geometry size. Again, we plan to work around this problem for computer model 3.5 GPUs, but it has to wait for other things to finish first.

Cheers,
Marcus

[FrankPooleFloating]

27M polys... I can almost wrap my head around that... but not quite. Of course, with stuff like
displacement maps, normals, instances etc.. I would never, ever, have a model anywhere even
remotely approaching those insane numbers of polys...

Man it has been quiet around these here parts lately.. hope some goodies (Octane goodies, that is)
are on the way sometime soon.

[roeland]

The maximum amount of triangles is about 26 million, due to a limitation in size of one of the textures used for storing geometry data. If you go over this limit, you will see the aforementioned error in the log.

Code: Select all

CUDA error 1 on device 0: One or more of the parameters passed were not within an acceptable range of values.
-> Failed to bind texture to device memory (data8)

--
Roeland

[pixelrush]

Interesting. I guess I didnt run up against the 2gb texture for geometry data limit with my 10million polys only cuda reserved space.
So potentially then Compute 3.5 supports 52 million polys for Octane ie. a 12 gb card alllows for 10gb vram for polys + 2gb for textures + film. Hmmmm.....mind you you would need 64 gb of system ram. Wonder if we will ever see a 12 gb geforce card Possibly then 40-45 million polys would be a practical max in the future..which is OK isn't it? I wonder if it would be possible to work with that size file in Blender or Max or even navigate the view port in Octane? Might be other limiting factors come into play...

[kavorka]

Also, If you do get this sorted out and then don't have enough VRAM to hold everything, enable just your Titan and disable your 680. That will increase the total amount of VRAM you have available.

[roeland]

The size limit didn't change for CUDA 3.5. At the moment most GPUs are limited by their amout of VRAM, to render 26 million triangles you need around 4GB of VRAM.

You don't have to disable GPUs to have more memory available, if one runs out of memory Octane will still render on the others. It displays the memory size available if you want to render on all GPUs.

--
Roeland

[pixelrush]

I admit to being slightly confused now. Abstrax seemed to say with compute 3.5 there was the potential for more than 2gb textures which apparently is where you store some data and that is the limiting factor in the number of polys...
Still 26 million is quite few unless you need to load 27 million.

[abstrax]

I admit to being slightly confused now. Abstrax seemed to say with compute 3.5 there was the potential for more than 2gb textures which apparently is where you store some data and that is the limiting factor in the number of polys...
Still 26 million is quite few unless you need to load 27 million.

With compute model 3.5 we would use a different system than textures and this way work around the limitation. The texture limitation per se would stay.

[oguzbir]

Thank you for the tips Roeland and Marcus,

Unfortunately for few projects I have to use the provided dense and complex meshes provided directly by the client.
It's now clear that I'm pushing the limits of octane . No not really. Will surely, get back to octane until this job is over.

So, nevermind.. Please continue improving things on your roadmap. This is a rare issue.

Best,