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
A few posts up is the resulting render after I deleted everything outside this one room, AND deleted any characters. So all that's left is the room I showed in the render, a few pieces of furniture, and even with that the GPU memory was 50% used (ie, 4GB).
And with Stonemason's city I loaded the entire thing into D|S recently, nothing hidden, and added 4 characters, and never had a render problem using the same GTX-1070. Never checked the memory usage, but I never had the slow renders.
And the point about just loading a sphere and plane requiring 2-3GB is really strange.
Are you leaving preview render on? That used to cause problems with memory a couple years ago. I don't know if it still does since I quit using it then.
If you're using most any V4 or Genesis character those textures are almost always 4K. You can easily check by clicking on one of its main texture files (e.g. torso) and looking at its Properties.
As Kevin notes, the full Stonemation urban sets have a lot of textures for the details. Load only what you need for the render.
Thanks, but Stonemason isn't the problem. As I said, I loaded his entire city with 4 characters and never had a problem.
Just now I reloaded his entire city with 2 G3's, and checked the GPU Memory Utilization, and it was only at 56% while rendering. I *assume* the big difference is that his textures are from long ago, prior to Iray. So maybe this memory issue is due to the Iray shaders consuming a lot more memory.
Though the core issue continues to be:
Why when you load a simple sphere and plane into D|S and render does it take 2.5GB of GPU memory?
Are you using an HDR to light the scene?
Richard,
Yeah, good point, I did have an HDR, but it was only a 3.6MB size image. And I changed my render settings to have no HDR, then re-started D|S and added just a plane and a sphere, checked there was no HDR, and rendered. And GPU Memory Utilization started at 6%, then went up to 30% during and after the render.
It doesn't on my machine. But I also don't use the same card for both rendering and monitor display, etc. There are too many variables to say why X happens in one instance, but not in another. I am also not using the latest bublic build, but still the last beta. There might have been a change, and if you still have another version of 4.9 on your machine (you can keep the public release and beta on at the same time), you can test if they differ. If the latest 4.9 is consuming more memory, you should send in a support ticket.
yeah, I guess that's my next step, trying with other versions.
I assumed that the issue of using the same card for rendering and monitors was a non-issue solely because it's driving the monitors before the render starts, but the memory is low at that point. Only when the render starts does the memory utilization go way up.
Windows 10 holds more card memory in reserve, if you're using 10, compared to Windows 7. The memory usage goes up when you start the render in part because of the calculations being done on the GPU. The calculations go to the CPU if you're out of memory.
It's better if you can keep the display separate from the GPU by using another card for it.
I have older stuff. I use a GTX 470 for my display and a 780 ti for rendering. It does OK for basic stuff.
Thanks Kevin. I get all that. The issue is why does such a tiny scene require 2.5GB of memory. That's it. When there's no way the total texture load comes even close, why is it using so much RAM? A 1920x1080 display requires something like 10's of MB. A 3k texture image also requires 10's of MB. But we're talking GIGABYTES. It just doesn't add up.
The scene starts off using CPU to get everything straight, then it shifts the render job over to the GPU. And my GPU has 8GB of memory. There's no way, in a tiny scene with a sphere and plane and no textures that it should require 2.5GB. Makes no sense.
BTW, here's some quick calculations on my guess for memory requirements:
So even if you add up a few monitors, and 20 big textures you might get around 500 MB.
But a scene with no characters, no big textures, and only a few monitors shouldn't be anywhere near 2.5GB unless I'm missing something.
Render resolution makes some difference btw, although not significant as a rule.
I don't see the same, and I use W10 pro; the 980ti is just used for rendering.
1 = I opened Gimp - before opening Gimp it was at 2MB
2 = Opened Daz studio - the default scene it loads has a couple of camers, a light and a plain in it.
3 = It didn't change when I loaded a small scene (figure, clothes hair and some props).
4 = Rendering said scene
EDIT:
Once the render finished, I started another render, leaving the previous one open. The other image shows RAM usage on the 980ti going up to 2681MB.
And the third image is with a resolution of 6000 x 8484. About 4000MB, so as I said previously image size makes a difference but not huge.
Personally, I'd trying rebuilding your scene; also try saving aspects of it as subsets and checking out what each one uses.
So nicstt, it looks like your results pretty much match mine? First image, #4, shows the memory usage as 2.2 GB while the small scene is rendering, correct? That's basically the same results I had.
Or am I missing something?
I don't have Windows 10. But the conclusion of linvanchene in this thread concluded "On a 1080Ti with 11 GB VRAM only around 9 GB are useable. 2 GB of VRAM are blocked by Windows 10 and are not available even though they are not actually used for anything." It was a thread also quoting Otoy and Nvidia. They were going to discuss the issue with Microsoft. https://direct.daz3d.com/forums/discussion/120541/windows-10-reserving-vram-update-gtx-1080-ti
Maybe it's something involving using a display card to render (I didn't read the whole thread) or an issue with Windows 10 and the 1000 series.
I rendered my own sphere and plane on my 780 ti in Windows 7 and it was minor usage.
AHA !!! So that's the story. How come you guys didn't tell me about this? Apparently W10 grabs a bunch of VRAM as some sort of safety net. And this has been an issue for a very long time. And apparently someone in that thread said D|S 4.9 used a LOT more than 4.8. I think they said an 8GB GPU used 1.6 previously, and 4.9 used an additional GB or something. Which seems to confirm what I saw.
And from my brief scan thru the postings, there ain't much you can do about it.
Well at least I'm not going crazy.
Well....not about this that is...
And I just checked the log file after starting D|S with no scene loaded, and it says:
"CUDA device 0 (GeForce GTX 1070): compute capability 6.1, 8 GiB total, 6.68421 GiB available, display attached"
So clearly at least 1.3 GB is being grabbed by Win10.
I think they also said that it doesn't really matter if you have displays connected, but maybe I'll plug in my old GPU and use that just for the displays.
Well, I did, but it got glossed over. Glad you're not going crazy! :)
Are you using Windows 10? It reserves an amount based on the amount available on the GPU (I think for a 8 GB card it reserves about 1GB!) Check with your GPU utility to see how much is in use before you start DS.
Textures take a HUGE amount of memory. a single 4k x 4k texure will be 64 MB uncompressed. Compression typically gives you about a 40% savings (but can be more or less, depending on the texture) so that's about 38 MB per 4k texture. If you have 4k textures used on multiple channels (diffuse, specular, normal/bump, displacement, opacity) each one will add as well. So a single mesh with 4k textures on all of those would be 190MB compressed. Iray also uses some of the VRAM for scratch space for the computations. It adds up FAST. If you are also using the same card for your display, the OpenGL stuff has to be in there too.....it can take up a lot as well. This is why it is best to have a separate card for the display, and render on a dedicated card.
As I understand i Windows 10t reserves enbough to supply each display that can be connected to the card, whethe or not a display is in fact connected. So a card with fewer potential connections would lose less RAM.
So let's say the fix is to get a second video card for Win10 and driving displays. How do we make sure Win10 reserves its VRAM in that card--not both?
I'm concerned about this now myself. Getting a second video card is a no-brainer, but I sure would like to know how to make sure it actually solves the problem.
Im using DS 4.9.4.115 on Win 7 and just made some renders with plane and sphere primitives. According to GPU-Z before render graphic card memory usage was 264 MB. During the render graphic card memory usage was 1311 MB which is around 1/3 (33 %) from 4 GB available. According to IRay render window texture memory was 0 and there was some little memory used for geometry and light.
For the second render I added some textures to the sphere and plane. According to IRay render window texture memory was 4 MiB. GPU-Z showed that for that render graphic card memory usage was 1314 MB.
Edited to add, I just restarted DS and rendered empty scene. According to GPU-Z graphic card memory usage was 1246 MB.
/shrugs
Yeh and I've had scenes take up roughtly the same on windows 8.0 or 10. I never checked what was available, so no clue if that was different.
Wow, for those of you trying to use dual GPU's, beware....
I'm working thru an issue that's been driving me nuts since yesterday. Due to concerns about having my GTX 1070 drive my 3 monitors, I figured I'd try re-installing my old GT 730 (useless POC) that came with my desktop, and run them both in parallel.
Bad idea.
When I rebooted it said it installed drivers for the card (which it should have had already), then rebooted. But before it started Windows 10 it hung. And hung. I waited for hours, but nothing. So I removed the GT 730 and tried again. Still hung.
Talked to my IT buddy and he said, oh yeah, we found that the older GT's don't play nice with the newer 10 series. So I took both cards out, plugged my monitors into the motherboard video HDMI, and WOW, it started Windows.
So I figured I'd put the GTX 1070 back in and it would be back to normal. No such luck. Still hung. So then I tried to do a restore point from yesterday before this mess started.
But for some reason it finished and said it encountered an error. Of course. The story of my life with Windows. Nothing ever works as it should.
So now I'm stuck with a computer that won't load windows unless I remove all GPU cards. And my Dell won't allow me to use the MB video plugs if I have a GPU card plugged in. Geez.
Anyway, be careful.
UPDATE:
For those who might be interested in issues you can encounter with dual GPU's if you're not careful...
The problem I described started basically because I installed an older GPU in parallel with my GTX 1070. That's it. That's all I did. And as a result I had to go thru 3 days of insanity, which included:
And finally I'm back where I was 3 days ago, with only a GTX 1070, and all my monitors plugged into that.
And my only mistake was plugging a second GPU into the PCI bus.
This kind of nonsense should have gone away 10 or 15 years ago. But we're still fighting DOS-era stuff. Amazing.
Nvidia control panel > PhysX > Processor: CPU, it took me months to find it.
Doing this, windows won't take anymore GPU ram and it will be all avaiable for DAZ, then it won't switch anymore to CPU. This solved all my problems.
The texture memory in the log is a buggy value do not trust it. The log counts a single texture as multiple textures if it is used on multiple materials. While iray counts it a single time. Also the log doesn't account for texture compression.
So always use a GPU utility to measure texture memory. Do not trust the log.
As others reported, the textures size is the main responsable for cpu reversing. You have to use the scene optimizer or resize the textures yourself. Personally I've found that without texture resizing most scenes are reversing to cpu specially on mid-range cards.
Empty scene, no lights, render setting set to scene only, tested with General Release and Beta, and it STILL uses a little over 2 gbs in the view-port render... we'll have to find a work-around, or get 12gb cards...
Iray implements texture compression so you can subtract about a 30% to 50% factor to that memory calculation. As for the 2GB overhead it seems to me that iray uses both a preempitive and lazy memory management. What this means is that it doesn't care of other applications and it handles the vram to maximize its own performances. That's why a configuration with one card for the viewport and one card for rendering is much more reliable than a shared card between the viewport and rendering.
In that scenario I guess the 2GB overhead may be a preempitive working space. That is, iray doesn't really need it for rendering, but given the 8GB memory space available it starts with 2G just to grab something. This behaviour may also vary with the iray version.
Alas I only have one GPU, funnily enough, I was playing Dragon age inquisition and THAT took only 2.6 gb of video memory...but yeah, thanks for the explanation all the same!