OcDS and Network Rendering - what is needed?

[birdovous]

OK, I tried to replicate the 2GB issue yesterday, but I failed as I was unable to get the >2GB export file generated. The most I got was ~1.8GB and if I tried to get more then all I achieved was to make the OcDS plugin or Octane Render engine itself crash still within the DAZ Studio as I exceeded the limits of 19M triangles and/or 32765 nodes

Unfortunately, these limits (or at least the triangle one) should be removed only after Octane v3 is released, which, if I understand it correctly, should have a feature called Out-of-Core Geometry enabling to get rid of that limit.

But GA of OC3 is still probably still some months away and OcDS v3... I don't dare to even guess...

[bepeg4d]

hi,
V3 is in alpha stage, so it's too early for speaking about it
About the 2GB limit, you should try to add to your 1.8GB scene a huge HDR or high res textures, should be enough, i think
ciao beppe

[gaazsi]

well honestly... my only problem with all this is that i invested 2k euro for a second rig for network rendering (and another standalone license for 300)

so i am not that happy about some workaround possibilities that dont even work for me atm and more or less wasted the money for nothing. time is money on rendering... and if i had to wait another 3-4 months until the next update where i can hope that it works, i maybe should better look to get familar with that "other" render program even if i like octane and especially t_3s fancy ocds features A LOT more

[bepeg4d]

I can feel your frustration
In waiting, I wonder if setting the master to be slave and vice versa has an impact on the result, unless is too difficult to do
ciao beppe

[birdovous]

hi,
V3 is in alpha stage, so it's too early for speaking about it
About the 2GB limit, you should try to add to your 1.8GB scene a huge HDR or high res textures, should be enough, i think
ciao beppe

OK did add a 4k HDR map and one more subdivided character with 4000x4000 diffuse, bump and specular maps on it and got to 2.2GB of the exported .orbx (export took little over 1 minute). Standalone 2.23 did open the file in about a minute as well without errors. Compiling the scene took about 2 minutes (maybe less) and network rendering started without any issues whatsoever.

So either the 2.2GB export file is still too small or we are facing more separate problems here.

EDIT:
I also forgot to add that the VRAM usage for this scene is over 4GB so it exceeds the 3GB limit of the 780 Ti in my slave box and apparently the Out of Core textures feature on the slave kicked in as the render runs at predicted speed with no visible degradation of the output or error messages on the slave.

[Spectralis]

Instead of network rendering I decided to go for two separate standalone and plugin licenses. Two system rendering works well for animation but obviously not for single images that need to share resources through a network. I can render the animation twice as fast with each system rendering half the animation.

[birdovous]

Instead of network rendering I decided to go for two separate standalone and plugin licenses. Two system rendering works well for animation but obviously not for single images that need to share resources through a network. I can render the animation twice as fast with each system rendering half the animation.

Hmm... That sounds odd. I would say the total time should be more/less equal due to the fact that the speed increase with GPU rendering is almost linear. But then again, I'm not doing animations so I have no real ground to base my assumption on

Plus the network rendering feature in the OcDS plugin does not work for me anyway...

[Spectralis]

Instead of network rendering I decided to go for two separate standalone and plugin licenses. Two system rendering works well for animation but obviously not for single images that need to share resources through a network. I can render the animation twice as fast with each system rendering half the animation.

Hmm... That sounds odd. I would say the total time should be more/less equal due to the fact that the speed increase with GPU rendering is almost linear. But then again, I'm not doing animations so I have no real ground to base my assumption on

Plus the network rendering feature in the OcDS plugin does not work for me anyway...

I render part of the animation on one PC and the other part on the other PC simultaneously so it's quicker than using just one PC. The older PC has a 780 installed so it seemed a waste not to use it somehow. It's about 30% quicker to render the whole animation with two PC's than with just one PC. A 40hr render now takes about 24hrs. The plugin price for the older PC was a lot cheaper than upgrading the GPU's on my new system. Networking would be easier and cheaper but apparently that isn't possible with OcDS atm.

[gaazsi]

i just wonder why it works for some and for others not...

[birdovous]

I can feel your frustration
In waiting, I wonder if setting the master to be slave and vice versa has an impact on the result, unless is too difficult to do
ciao beppe

OK I just tried to switch the master and slave around. The only difference is, that the previous master (now slave network node) is visible in DAZ immediately and also immediately switches to ON state. That however, is about it. The problem with the rendered output being desaturated (and thus totally unusable) still remains, no matter who is the master and who is the slave.