Hi,
I'm running into a few issues with Mantra and tiled rendering and wondering if anyone can provide some assistance. I'm trying to render some highres arch viz stills for billboard usage (9600*3600 pixels) and finding that the tiled render option on the Mantra ROP is not giving the behaviour I'd expect.
Render times for the frames on a single machine are very high (~70 - 80 hrs , so the only way I can see me being able to render these promptly is to split up into chunks and distribute across the farm. The problem is that I'm not noticing render times or RAM usage for the chunks being proportionally lower (given tile size) than the original full frame render. So if I split into 100 tiles, the render time for each may only drop to around 25% of that of the full frame and the RAM usage stays approximately the same, meaning that doing a tiled render will actually take much longer than using a single machine). This isn't the behaviour that I'd expect as I would have assumed that the render time would be substantially lower, maybe not 100th, of the full frame but certainly faster than I am achieving. I would also expect the RAM usage to be much lower, decreasing in line with the resolution of the tile size.
I've got a workaround using multiple cameras using the crop parameters, and ROP's, set to render at an appropriate fraction of the resolution, but seems that this is how tiled rendering “should” work?
Is there something obvious and fundamental that I'm missing with regards to rendering tiles in mantra, or is this behaviour to be expected?
Tom
Tiled render issues
3614 5 0- IP_Animation
- Member
- 8 posts
- Joined: Nov. 2014
- Offline
- jarenas
- Member
- 146 posts
- Joined: Jan. 2018
- Offline
Hi:
I found this old post and I'm finding a similar issue with Mantra when rendering very large images (8000*6000), with RAM usages of 80-100GB for a single render, and of course, having only 64GB of RAM, the render never ends in a reasonable time. I've tried tiled rendering, but RAM usage is not much better…
Is this RAM usage reasonable for a simple test image with just one displaced 8K material and a hdri 12K background?
Excuse me if this question is extremely basic, but I'm just starting to learn Mantra (precisely for future large size rendering works)..
Thanks in advance!
I found this old post and I'm finding a similar issue with Mantra when rendering very large images (8000*6000), with RAM usages of 80-100GB for a single render, and of course, having only 64GB of RAM, the render never ends in a reasonable time. I've tried tiled rendering, but RAM usage is not much better…
Is this RAM usage reasonable for a simple test image with just one displaced 8K material and a hdri 12K background?
Excuse me if this question is extremely basic, but I'm just starting to learn Mantra (precisely for future large size rendering works)..
Thanks in advance!
- dlee
- Staff
- 420 posts
- Joined: Sept. 2016
- Offline
- jsmack
- Member
- 7762 posts
- Joined: Sept. 2011
- Offline
- dlee
- Staff
- 420 posts
- Joined: Sept. 2016
- Offline
jsmack
Does mantra's ‘offscreen quality’ consider the area outside a tile to be ‘offscreen’? If not, then I would not expect to see any memory drop from using tiles, as the dicing requirement is unchanged from the full frame.
I don't believe so - it should be based on camera fov only. After all, we wouldn't want rendered result to be any different when doing tiled render vs non-tiled (or differently tiled) renders.
- tamte
- Member
- 8551 posts
- Joined: July 2007
- Online
IP_Animation
I've got a workaround using multiple cameras using the crop parameters, and ROP's, set to render at an appropriate fraction of the resolution, but seems that this is how tiled rendering “should” work?
so are you saying that rendering a Tile takes different amount of time and ram than rendering the same region by using Crop settings? if so it sounds like a bug as I'd expect both approaches to essentially do the same
Edited by tamte - April 20, 2018 17:35:27
Tomas Slancik
FX Supervisor
Method Studios, NY
FX Supervisor
Method Studios, NY
-
- Quick Links