iRay falls back to CPU (in big scenes)
surody
Posts: 261
No matter what I do the renderer always falls back to CPU if the scene is pretty big. I turned off SLI and I also tried using only one GPU. I tested Daz 4.10 and Daz 4.12 and both have the same issue. Nvidia driver 430+
Daz 3D is part of 
Connect
DAZ Productions, Inc.
7533 S Center View Ct #4664
West Jordan, UT 84084
Licensing Agreement | Terms of Service | Privacy Policy | EULA
© 2025 Daz Productions Inc. All Rights Reserved.
Comments
You must be using up all the vram, it's the major setback of using iray. You need to fit it into the vram of your card, which gets used pretty fast with all the 4k+ texture maps in people and scenes. There is some tools that can help mitigate it, https://www.daz3d.com/scene-optimizer being the best one I know of. Even with that though, in big renders I end up having to do at least 3 different render passes(foreground, midground and backround usually), and spend a bunch ot time compositing them together. It's a real bummer lol.
This makes the Titan RTX pretty tasty for me tbh
Wish I could afford a beast like that lol. Stuck with a 1070 and 960 to render with for now
Just use the scene optimizer addon as @TheKD suggested, this also makes the render times faster so it's a win win. Then rendering a static background for composition is also a big saver for animation.
If the scene is eatting too much Vram, the iray renderer goes to CPU, if you want to increase the vram above 12GB, the only options are Quadros or RTX cards in NVlink mode that adds system GPU resources together.
The only other option is to use the Scene Optimiser to tweak things to fit in to the avaiable vram.
Does NVlink actually stack VRAM?
If you have Nivdia Quadro's with the right NVlink mode set it does stack the vram.
Video comparision of SLI vs NVLink on RTX and Quadros and NVLink modes.
The Cheapest option is just to use is Scene Optimiser. and reduce texture sizes and mesh detail for further objects in the scene.
The only other option is a Nvidia TITAN RTX with 24GB of Vram.
You can try this as well - https://www.daz3d.com/resource-saver-shaders-collection-for-iray
Info on the promos was from DS 4.10
Thank you, I already own this product.
You're welcome!
Aside from textures, have you gone through the scene and hidden or deleted every item that is not seen by the camera or in reflections?
Have you shut off all the nodes on the figures that are not at all seen?
Have you checked how many items are set to Hi-Res and what their Sub-D settings are?
Large scenes for me are composites. I render background in 3delight and save png or tiff. Next I render main characters in Iray by themselves. Composite in GIMP or photoshop. I have the optimazation tools sold here but it is too much work.
Yes, I usually do that. But when I want to render a scene showing a lot (even while using instances) it may happen that VRAM is easily exceeded.
Instances can still tax a system, especially if there are a lot of them. Try setting Instancing Optimization to Memory, (in Render Settings > Optimization.) I just tested this on Urban Sprawl 2, which I believe doesn't use any instances, and still saw an improvement in the amount of RAM used by my GPU, (using Z-GPU to monitor.)
Something to keep in mind, though, using the Memory setting can have unwanted results on some objects, especially skin.
It really depends on how much time you want to spend preparing the scene vs how much time rendering in Iray will really save you. If you can render overnight, and the image is completed by the time you get up the next day, you many want to do fewer things to prepare the image. But if it's taking days, you'll probably want to utilize all the tips from this thread.
I'm currently testing Iray Section Plane Nodes and found very little benefit with Urban Sprawl 2, but it's a Stonemason set, and I expect it has optimized materials. (Pre-Iray. I converted all materials using the Iray Uber Base.)
I'm going to add some high-rez, high poly props to the scene and see if the section planes make an appreciable difference. I'll let you know here, if they do.
ETA: Sadly, it didn't make a difference. I added a couple of characters with hair and clothing, and a helicopter, all out of frame and behind the Iray Section Plane Nodes. The load on the GPU's RAM was the virtually the same, whether or not the planes were active.
I was very slow in switching from 410 to 411 and things went well for a couple days when I hauled out my big scene and moved to another section of it to render and wham into cpu.
----
so went back checked everything including closing anything that might have been using vram and went back to the original camera and it rendered just fine..
despite the fact it was using 12090m of ram... glad the last card upgrade was a used titanX ... one hour and one minute
----
but I knew I hadn't seen messages with that much vram usage ... so opened the same file back in 410 and the memory usage maxed at 10800m and the render only took 50 minutes about 15% faster. (file was closed to flush the card so starting from scratch just like the other time.
As it happened the green enamal paint I used on the light poles was very reflective and it reflected the image of the lamp on every other lamppost.
So I beat on the shader until it agreed to lie there and do nothing and just be green, it's easy I told it.
Anyway redid the render in 410 wanted to see if cutting out all those extra little reflections would help
Okay this time the vram went to 11900m .. but it did knock off three minutes to bring the time down to 47 minutes.
---
This is the same scene file with the same camera, three different memory usages and three different times.. but 410 was faster and used less memory.
---
but out of what 411 wanted to have a 12g file 410 ran the same file a little closer to the bone and it probably would go on an 11g card.
----
and looks like I shaved off time by changing the paint to avoid all the little light reflections on the poles.
Scene optimiser is pretty amazing. Ive done more than a few renders of 9 complex daz characters with lots of clothes and very high polygon hair .. in complex scenes. On a 1070. I rarely have problems (now).. heres how.
First my monitors are hooked up to my 1060 and the 1070 is only used for rendering. Next if im having problems i delete anything you cant see. If i have a plate out of camera.. i delete it. And last i use Scene optimizer. It's pretty amazing and i SHOULD use it in all my renders. You can completely control every imagineable parameter and i dont think ive ever noticed a difference (i usually only downsample scenery and some clothes.. never a model). In some scenes literally a tiny number of textures can take up Massive amounts of ram if theyre repeated enough
Lovin' that Dick Tracy-esque scene!
Yes, but only presently with Iray Server running on a Linux system (current versions of Windows and OS X have incompatabilities with Nvidia's official underlying memory management scheme.) See here for all the juicy details.
Another option, besides reducing texture resolution, for optimising textures for multiple characters is to use the same maps for certain things. I tend to have all my characters use the same mouth, teeth, iris, pupil and sclera maps (UVs are cross gender compatible for G8). This also means not having to adapt these materials to look consisent across the scene when using multiple character materials. Another option is to have all, say, female characters use the same spec, bump, normal maps and just change the diffuse textures (this might not be an option for close ups).
Additionally, some prop sets are really heavy on resources - rougness mats, metallicity maps (including completely black mats on occasion... why??? /rant) etc in addition to the regular ones. For a background item you can decide if this is really necessary - are you using depth of field? is all that detail actually visible? Can you replace some textures with tileable textures from a texture pack (or somewhere like texturehaven dot com). Depending on your answers, you might benefit from removing some of these materials and saving the prop/s as an optimised scene sub-set.