So, how does the OOC rendering work anyway? Tweet

[birdovous]

Yesterday I was rendering a project that took over 7GB of VRAM and I thought that having Out of Core rendering enabled and enough physical memory assigned to it should do the trick.

Apparently, OOC either does not work, or I did something wrong, or I misunderstand this feature.

What I thought until yesterday, is that when the GPU does not have enough of physical VRAM then it should be able to use system RAM allocated to it and being able to render such scene at reduced speed.

I'm using network rendering with main workstation running DAZ and having 2 Titans X (with 12GB VRAM each). Then I have a network Octane node with a legacy GTX Titan (6GB VRAM) and GTX 780 (3GB VRAM). Both the main workstation and the node have OOC enabled and 8GB of system RAM assigned to it.

However, with the scene I was rendering yesterday both Titans X were crunching away happily as they could fit the scene into their physical VRAM, but the network node crashed because Out Of Memory.

Could anyone enlighten me on this? Is it a bug, or the OOC simply does not work like this? And if this is the case, then how does it work? What has to be done for this feature to kick in?

Thanks for any info.

[rappet]

How large is the scene (on the master) after having set OOC?
it should be smaller then your smallest Vram on the slave your are running.
And I assume on the slave you did set the OOC in the Octane Slave Daemon window box, right?

[birdovous]

How large is the scene (on the master) after having set OOC?
it should be smaller then your smallest Vram on the slave your are running.
And I assume on the slave you did set the OOC in the Octane Slave Daemon window box, right?

Yes when the slave starts it has OOC enabled and 8GB of system RAM assigned to it.

Wait... the scene has to be smaller than the smallest VRAM on a slave?! Then I probably do not understand the feature at all. I always thought that OOC should enable you to render scenes that are actually bigger than the VRAM you have physically available on given GPU?

The master node has 2 Titans X with12GB of VRAM each. The slave has a legacy Titan with 6GB of VRAM and a GTX 780Ti with 3GB of VRAM. The scene itself was something over 7GB (as shown in the DAZ plugin GUI in VRAM utilization).

[rappet]

OOC does save some VRAM space on your GPU, but still then the scene ofcourse needs vram.

If after OOC your scene uses 7 gb on your master you can render it on your master with your Titanx having 12 gb available.
But your slve has 6gb and 3gb, so it will not able to load the scene.

example on your master:
Suppose you have Titan X and a 780 (3gb) and both are activated.
If the vram needs to load more then 3gb for the scene then the Titan X won't run either...
then you have to de-activate the gpu's that can not handle the scene to make it possible for the otheres to run.

Same goes for slaves.
i.e. if your scene needs 5 gb of vram, you rmaster will run it ofcourse and your gpu's with 6gb can load the scene.
But if you have on gpu with 3gb availbale it won't run if that gpu is activated also the other gpu's (6gb) won't run, because all activated gpu needs to be loaded the same scene.
You can disabvle only the gpu with 3 gb, and the other will run that scene again.
But to disable just one or two on your slave it is not so easy in the daemon slave window when starting up, coz'it is not easy to know which gpu is which.

I hope now you understand how it works.

cheers

[face_off]

As Jeroen says, the 3gb card is the problem. Try getting the OOC settings working on a single card first, then add additional card on that PC, then finally add cards from network rendering PC's in order to see which step causes you a problem.

Paul

[birdovous]

Well.... guys... now I think I'm lost, or I totally fail to understand the purpose of OOC. I always thought it should enable cards with not enough physical VRAM to render scenes that do not fit there at the expense of speed?

Like geometry gets in the VRAM, textures in the normal RAM allocated via OOC feature... ?

Truth be told, so far, I have never seen the OOC being used although I have it enabled on both the master in DAZ plugin and slave in the Octane network node... And from what you both say, it looks like that I never even will...

[Zay]

Export the scene to orbx. Open the orbx in Octane Standalone on the slave machine and see if the cards can handle the scene.

[birdovous]

Export the scene to orbx. Open the orbx in Octane Standalone on the slave machine and see if the cards can handle the scene.

My Titans in the main workstation can render the scene without problems as they have 12GB of VRAM each and the complete scene has slightly over 7GB.

I was posting this thread because it seems that I fail to understand the purpose of OOC since I always thought that it should enable cards with low VRAM to tackle bigger scenes. And as it seems I was mistaken all the time

And to make things even worse, now I have absolutely no clue what is the OOC feature actually good for? (apart from the user being able to turn it on or off)

[rappet]

OOC is not that complicated... When OOC enabled, textured are stored outside vram and therefor you might be able to have scenes rendered which otherwise are too big to be loaded into the vram.
I.e. A scene is just little too big for 6 gb vram ...and you are not able to render with Titan 6gb card
Then enable ooc and the scene stores textures outside vram and your scene (your model in numbers of triangles, the resolution also takes effect on loaded vram, but then not the textures that are loaded outercore) might use i.e. 5.7 gb for the vram and therefor you are able to render.
Ooc has downside (I thought it would be slower in loading or something like that), but to be honest I never used it ... Yet ...

I hope you understand now