Workshop

Made for webinar
Render by RTX (non path tracing)
2k resulution fps:35-44
720 resulution fps: 100+



2 Likes

Looking good, man!

Very nice!

What GPU are you using?

we test on rtx 3080ti and rtx 3090
CPU: ryzen 3970
Mother bourd: rog H2F TRX40
Memory : 128G
Driver:Samsung 970 pro

3080ti 2k resulution fps:31-39
3090 2k resulution fps:35-44

project did not Optimization.no refrence usd, no instance,no reused texture.
its a preproduct test.

We Mainly using 4K res texture for testing Connect plugin io limit.
We use 2 day to build this scene. ( but 25 day adjust rtx gi)

but the trouble is nuclus workstation version unstable.
even I know workstation version isnt support cache feature, we get a big io limit when we try to import some asset from nuclus.

the thing is, when we try to import like 70mb asset from nuclus to create, most time create will instancely crash.
at that moment i think maybe the issue come from my computer driver or graph card.
but unfortunately , they are no trouble.
then, we focus on Ethernet io data size. but the maximum data stream size is 40MB(but asset size are 70MB)
for now, i believe the trouble are coming from create buffer.

there are 2 way get crash from create.

1: when you drag multiple asset from nuclus or when you drag something into scene but you the asset size too much big or when you drag something but sadlly you are intersect something from scene.
i think this is a io bug.

2:when we import aseet but the texture size are big. and shader maybe binding 7 different 4k texture. at that moment, create loading all texture in once quest and also create starting render this thing, there are no buffer to handle both big io and the io size over then cache size. so it crash (i guess ).

but we already get a nice test.
this project we separate 5 difference usd layer for teamwork.
floor, wall, brick,etc are procedural from Houdini.
model asset we use blender export usd (but when we finish job, you guy public blender 3.0, unlucky.)
assemble and modify asset from maya.
lookdev and r&d built-in Houdini.
texture on substance painter but we using our substance painter livelink plugin( but we setting 10second automatic send data to omniverse create,each time we sending data,the data size roughly like 300mb, but create didnt crash, confused …)

for 25 day job we focus on how to adjust rtx render mode effect closing raytrcing render mode, now, rtx render closing 80% pt render.

next project we will focus on high quality character ,inclouding hair dynamic test, cloth dynamic test, animation and blendshape test.
hopefully we can show more awesome work.

thanks nvidia.

1 Like

Hi there,

It sounds like you might be running out of memory due to the large textures causing the crashes.

We have several optimizations in the works that might help. Texture streaming is on the way, as is some fixes to release memory more effectively.

Can you tell me the size of the textures you’re trying to load? Do you see “out of memory” in the log files at all?

I mean, the limit maybe coming from nuclus servers, cause most crash time was when i drag some asset from nuclus, but when i drag from driver it work well.

Looks great!