Oh OK then that could be a fault of how my system is setup that it behaves poorly at it’s limits causing a crash that on a better tuned engine revving to the redline means going faster not catastrophic failure
Exactly. Did you disable the second 3090? I want you to look at that as a factor as well. I really don’t like that PCI Express warning. That is pretty serious.
1 Like
I can try that pretty easily with a CLI argument let me see if I can fit that in during my morning routine
No I do not mean disable it just for Omniverse Kit. I mean DISABLE IT totally in the system. Remove it either physically or at the device root level.
Yeah totally overkill but sure. I could also fiddle with the slots and bios and see if both cards are on either pcie 3 or 4 not both
Well I guess you could start at the system level, before physically trying to change the cards. But it could be a factor
1 Like
https://www.asrockrack.com/general/productdetail.asp?Model=ROMED8-2T#Specifications This is the mother board which appears to be all PCIe 4.0
Also something I want to show you. Whilst AOVs can be cool and useful, they massively reduce the rendering performance because writing all of that data and working with multiple layers has a big impact on CPU and disk.
Video 1 shows rendering with AOVs. Look at the GPU usage, very low. around 30%. Render frame time is about 1.5 mins
Video 2 shows the same rendering but with AOVs off. Look at the GPU usage, now at 80%. So almost 3 times as fast. Which means way less rendering time. Render frame time is now just 15 seconds !!
So I would honestly only render AOVs if you HAVE to have them. You should really be rendering with Realtime and now Realtime 2.0
The NVLink bridge might be the issue I won’t have time to explore that this morning before work
Yeah I have seen the inefficient resource usage but the depth pass is a requirement perhaps getting it out of the debug pass or override shader is my path to glory
Now here is the power of realtime. SO much faster and much much cleaner. I always use realtime. For the project you are doing, based on what the style looks like, lighting, lack of complex refractions and transparencies, you could totally render this in realtime, which basically means less that a 1 second a frame. No more rendering basically.
It was so fast I had to render it TWICE to capture it !! LOL
1 Like
Thanks for all the information as they say “the obstacles show the clear path”
Just so you know, REALTIME does great depth of field if you want to do it “in camera”
1 Like
Oh I just saw this !!! Actually we strongly recommend not using NVLINK !! This is not a good idea. You have a pair of 3090s with NVLINK? That is pretty rare. That was the last GeForce card we at NVIDIA did with NVLINK. It is now discontinued.
Please consider strongly taking off the NVLINK. Omniverse Kit cannot use it. We do NOT share GPU memory across GPUs like this. This may be making your system very unstable.
1 Like
yeah good idea , I think the PCIe 4.0 bus makes it fast enough that it is kinda a moot point anyway, but fan boys will be fan boys.
Thanks for the advice.
WE MAY HAVE A SOLUTION!!
in my Machinima backwater where the application of this technology solves a real world problem.
It seems using 16bit half float OpenEXR files with RLE encoding, I can get about 70% GPU Usage AND get a sequence of PtZDepth ldrColor and hdrColor which are not corrupt.
I am really sorry about locking horns with you on this thread, you really were trying to help me out.
Its funny we were both right and in opposition
its much better to be both right and in collaboration.
This is what I am talking about
HUZZAH!!!
Just checking in. Hope everything is good with you. If you have any more issues please open a new thread. Happy to help.
This is what I got, didn’t end up using the depth pass, but its good to know I can get it if needed.
This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.