I´m having problems rendering octane render elements via the 3dsmax render elements in high resolution.
The scene has about 7 Million triangles and I want to render it in 8000 x 6000 with two Titans.
I can render it in this resolution with no problems using about 3,5 Ram of GPU Memory.
I wanted to get Ambient Occlusion, Direct indirect Reflection, three Light passes, Zdepth and object and material ID with my Beauty rendering
and added them in the 3dsmax render elements dialog.
But the render always crashes (Warning: render has failed) before even showing me the GPU Memory used.
I tried the CPU memory use up to 10 GB but it makes no difference.
Even if I remove all render elements exept the object ID pass (which shouldn´t be to costly) I cannot render the image in the 8k resolution.
With alle elements active it works fine up to 3000k.
So I was wondering: How expensive are the render elements Memory - wise.? Can I add the memory needed for the whole scene for each render element?
Why doesn´t CPU memory usage help? Is this only used for storing textures?
Seems to me you can´t get your render elements in one render with your beauty pass in higher resolutions with even an 6gb Titan?
Memory consumption of render elements
Forum rules
Please post only in English in this subforum. For alternate language discussion please go here http://render.otoy.com/forum/viewforum.php?f=18
Please post only in English in this subforum. For alternate language discussion please go here http://render.otoy.com/forum/viewforum.php?f=18
- stratified
- Posts: 945
- Joined: Wed Aug 15, 2012 6:32 am
- Location: Auckland, New Zealand
Hi,
Each render pass requires a separate buffer on the GPU - except for the info passes which share the same buffer. Each buffer is in HDR format, 4 bytes per channel and 3 or 4 channels per pixel depending on whether you use alpha or not. So for your scene, you would need (assuming alpha)
What you can do is disable at least 3 render passes or you can render in tiles (offset via lens shift) and stitch them together in post.
We plan to have a solution for this in Octane 3.0.
cheers,
Thomas
Each render pass requires a separate buffer on the GPU - except for the info passes which share the same buffer. Each buffer is in HDR format, 4 bytes per channel and 3 or 4 channels per pixel depending on whether you use alpha or not. So for your scene, you would need (assuming alpha)
4 * 4 * 8000 * 6000 = 768000000B = 732MB
. You will need 6 buffers which gives you 6 * 732MB = 4392MB
. You would need at least 4392 + 3500 = 7892MB
on your GPU. CPU memory doesn't make a difference.What you can do is disable at least 3 render passes or you can render in tiles (offset via lens shift) and stitch them together in post.
We plan to have a solution for this in Octane 3.0.
cheers,
Thomas
Thank you for you answer and explanation. 
Good to know CPU Memory doesn´t help here.
You suggest to disable at least 3 render passes. But the thing is, I can´t render the scene with even one render pass activated
Not even with only Object - or Material ID.
When there are now render elements at all it renders fine.

Good to know CPU Memory doesn´t help here.
You suggest to disable at least 3 render passes. But the thing is, I can´t render the scene with even one render pass activated
Not even with only Object - or Material ID.
When there are now render elements at all it renders fine.
corei7-4930k 6 x 4,0 Ghz, 32GB RAM, 2x 980 Ti
Win7 64bit, 3dsmax 2014 / 2015
Win7 64bit, 3dsmax 2014 / 2015
- stratified
- Posts: 945
- Joined: Wed Aug 15, 2012 6:32 am
- Location: Auckland, New Zealand
That's odd - so when you enable 1 render pass on top of the main pass it goes out-of-memory? That should not happen. Is there something in the log window when this happens?DIO wrote:Thank you for you answer and explanation.
Good to know CPU Memory doesn´t help here.
You suggest to disable at least 3 render passes. But the thing is, I can´t render the scene with even one render pass activated
Not even with only Object - or Material ID.
When there are now render elements at all it renders fine.
cheers,
Thomas