Show Us Your Iray Renders. Part III
This discussion has been closed.
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
Remember that video cards can work on a type of compressed texture. When I program pixel shaders in OpenGL or DirectX, I send the texture to the video card in DXT5 format (or let the video card/driver compress it for you). This gives a 4:1 compression ratio. If you don't have too many sharp, abrupt changes, you shouldn't notice any compression artifacts. I wonder if iRay makes use of this. I'd be surprised if it didn't.
It is not about if I believe or not, facts are facts and it should not fit , in the other pic all your loaded was 88 genesis with only one texture set and you could go for 300 of them this way , this no matter, the question was only if each of the model has separate texture set and not duplicated and that will make big difference , this no matter you use JPG PNG or TIFF it is about the pixel not the format .
so compression will only slow down the render time , you know that rendering using tga format textures will be faster than jpg ?
Your scene may be around 4GB from what I see but it is not possible to have 12 different texture sets on the 12 models unless it was reduced in resolution so I am going to investigate it more deeply by doing another test
and then will let you know , I had my info direct from the iray source so nothing I imagine or read in this forum and I did not found any info about Iray handling textures different as it did before .. and if it does then it re scale it down what explain everything but no documentation about that so we can just assume ... but after my test I will be sure what's going on
Mec, your math seems to be fine, but you have made errors in how Iray handles scenes. My testing and real world reality are very different from that simple math. My best guess for this is Iray is a lot more masterful at handling texture files than anything I've used before. It is not a matter of just multiplying out some numbers.
Attached is a re-render of that scene... Sorry it's just a quicky. I didn't save the original. Also attached is a screen shot of GPU-Z as the render was in progress. Clearly, this 4GB GTX 980 was functioning and under 4GB. In fact both of my 980s were running through the entire render process while my 2GB GTX660s dropped out, as expected. I keep CPU turned off, which is not a suggested practice unless you are really on your game with your card(s). I do apologize for saying "well under 4GB VRAM", as it is closer than I remembered. Sorry, it has been several months since I ran that test.
I attached another Iray test render with 88 figures in it. This was done on my laptop using a 4GB GTX780M. I did this so long ago, I have no idea about VRAM use. All I do know, is it positively rendered on the card, which was the point of this test. Yeah... ugly but impressive that it worked.
Also, a lot can be done to optimize texture files. First, some can be gained by losslessly saving them or getting rid of where text can be saved within a jpg. Of course, carefully choosing a proper quality level is important, but after that, there are programs that will re-optimize the Hoffman compression with 0 loss to image quality while cutting file sizes by around 1/3rd. It's a new game. I've been working very hard on texture conservation with so far, great success.
I sure hope you believe me this time as I do not plan to do a video to show this as it happens. :) And if you do now believe me, your apology is accepted.
BTW, you are getting some great skin renders. Good work!
All I know that Iray use memory per pixel and there is no info about working on compressed textures rather decompressing them before usage ending in slower calculation time , I tested with Tga and Jpg and tga was faster , reducing speed from 1.30 min to 32 sec with uncompressed textures
Yeah, I forgot to mention that about the warranty, too...
Something more to be said about progress then, eh? :-)
So far, testing for Iray is showing almost an exact linear line on speed of renders based on the number of CUDA cores. My 2 GTX 980s are about 25% faster than my one Titan X Overclocked. The question then becomes fitting the scene onto the card. Two GTX980s cost the same as one Titan X and 'if' the scene fits, your rendering will be faster. I have so far tested GTX660, GTX980, Titan X and even lowly oldie GTS450, which is something special as it doesn't drop out like it should. Times based on CUDA core count have been extremely close to linear. This has also been confirmed by another PA friend of mine and his testing has included some other cards with still the same end result. Money does not equal render speed. CUDA core count does. Reviewing Sickle's test render thread seems to show the same results.
That said, it looks like a good high end card for almost any scene could be the GTX980Ti. It should perform almost as fast as the Titan X and cost $400 or so less. It does only have 6GB VRAM instead of 12GB. I know some people who have gone over 6GB with some of those insanely busy scenes... can you say Totte? But heck, he's insane anyway and will always go beyond the limits! For me, the scenes I have done so far that got close to 4GB have quite frankly been a pain in Studio and OpenGL trying to move them about in spite of being in Smooth Texture mode.
There are no hard rules here. Some products are going to be more efficient than others. The bottom line is to pick a card with enough VRAM to hold the scene and 4GB is a very good point for doing that without breaking the bank. Then add as much as you can afford, to buy as many CUDA cores as you can get. A 500 CUDA core 6GB card is not going to be particularly fast. I'm operating now at just under 10,000 CUDA cores and for many scenes, I'm just about to move around with the viewport in NVIDIA mode in pure realtime, or at least for smaller scenes.
At the moment, I believe the GTX980Ti is going to be a very sweet spot. 6GB is nice and almost 3000 CUDA cores. 2 of those is just a bit more than one Titan X and almost twice as fast. The one test I saw run by Spooky, showed like linear performance on his I believe it was a K6000? It's over on the test render thread somewhere. The more expensive K6000 is a much higher cost per CUDA core than other solutions.
Anyway, hope that helps. This is what I've seen doing careful timing on real Iray renders... some of the card specs can go out the window unless you also want a really awesome gaming computer. :)
...I'm up there with Totte as well.
It's too bad that there isn't some way that some enterprising hardware engineer can find to make VRAM cumulative. Then dual 980s would be an excellent solution. However, until that happens (if it even can) the Titan X will remain the "holy grail" of GPUs for us. At least at 1/5th the cost of a Quadro M6000, it is somewhat reachable.
Yes, that was a disappointment for me. I suppose there is good reason. As fast as they are, I guess the ram needs to be close to the processor. Yes, the speed of light matters! The Titan Z looks really great, until you realize it splits the 12GB VRAM between its two processors.
And yes, some day I will most likely blow past the 4GB and my next card I think will be a 980ti... and I'll likely blow past that 6GB, which is why with great pains I bought the one Titan X. So it can be the final fall back for me in what are rare situations for me.
Also, Texture Atlas can be your friend. We really don't need hirez textures for distant figures or props. There's lots of ways around this.
...yeah, however, choosing which textures to lower resolution on and doing so is fairly tedious work.that impacts the workflow. Wish I had enough GPU cores to constantly work in Iray view mode. Sadly that's about a 1,300$ proposition (based on two 980TIs) Spooky seems to be able to do it with dual Quadro K6000s (total of 5760 cores). Dual 980 TIs will give you 5632 - not far off..
Of course dual Titan-Xs will give you 6144. (hmm for an MB with 4 PCI 3.0 x16 slots = 12288 cores - all for less than the price of a single M6000) .
Watch in two or three years they'll be rolling out a the "Einstein" technology "Titan Colossus" with 18 or even 24 GB of video memory and 5,000 CUDA Cores. :cheese:
Just posting this. Let me know what you think please.
...are you using an UberVolume cone or cone primitive for the front light? It looks rather solid as the light is actually shining on it.
Ok I tested .. 2 genesis female with different texture sets 2006 MB VRAM, then added CPU the memory was only used by half on the Graphic card the rest I am sure virtual graphic memory via pci , then added 3 more genesis total 5 with different texture sets .. the graphic card memory reached 2GB
so with 4 GB VRAM and CPU you can get 10 genesis textured in the scene or use Atlas for more than 5
but on your card only you can do just 4 with full different textures with 4GB or it will starting to render very slow
Bellow how much memory was loaded when I render scene with 2 full textured genesis with different textures
Iray use not only VRAM but also virtual video memory, if there is not enough on the card it handle the rest via pci but resulting in slower calculation and rendering and if there is not enough , it will crash .
so as I suspected before the card can handle so and so for full performance before it get into slower mode , I am glad I choice the 12 GB as I am going to have a lot of fun working with it and not worry about reducing everything as I am doing it for the last couple of years and I have enough of that workflow
Just playing around with the beam prop that came with the sentry drone model. I converted it to frosty glass shader from the iray selections. I could up the opacity for it a bit.
Let me know if you have any tips to try.
...don't have that model so didn't realise it came with a beam prop. The other two beams from the sides look fine, did you make any changes to them?
I finally finished my first iray render!!
http://drowelfmorwen.deviantart.com/art/Meditation-for-the-Huntress-Iray-537137997
Also, does anyone know how to make things glow/emit their own light for iray?
Use the Emissive Iray Shader
sadcubes, thank you! oh my goodness, I even saw this shader a dozen times *slaps forehead*
I'm sorry but this makes no sense to me? Some of what you wrote seems to contradict itself? Perhaps there are some typos? It seems that you have concluded that only 4 figures with different skin textures can be loaded onto a 4GB VRAM card. That is absolutely wrong. Now, if you chose skin sets with maybe 8k material files with maybe additional 8k files for things like nails, you might be able to accomplish filling 4GB of VRAM. However, this is not the norm and I believe you are misleading the community following this thread. Or perhaps there is a misunderstanding by me, which is totally possible as I am left confused by your post, or maybe a misunderstanding by you about your system.
Is the GTX 760 your only card? If so, that is a 2GB card so you have no real world data on a 4GB card. Perhaps you are just doubling what your 2GB card can do? If so, that is again bad math. The card and the system can easily need 1 GB VRAM leaving you only 1 GB for your scene. So a 4GB card can hold "very generally speaking", a scene around 3 times the size. How much VRAM was in use before you started the render? Add to that about 500mb if you have other common things running such as a web browser and/or email program. Both use video resources and even OpenGL. When I've crashed my cards, there was normally an OpenGL error notice and lots of white windows left around. If Photoshop happens to be running, I could almost forget about rendering on a 2GB card. What resolution is your monitor. If it is a high rez monitor, that takes more VRAM.
Could you please explain what you mean by 'slow down'? You say the scene was 2006 MB VRAM then say only used half on the Graphic card? I'm lost there.
I found I could only reliably run a 2GB card with 1.5 GB showing in GPU-Z, if it was also the main card used by the system, or in other words the card which has the monitor plugged into it. That extra half gig is needed by the system for spikes that happen faster than a utility can record them. If you have all other applications closed at the time, and if on Windows, switch out of Aero mode, get rid of any desktop background image, you can both clear some VRAM and push that up fairly reliably to 1.7gb or 1.8gb in use. DAZ recommends a 4GB card because really, only one figure is 'reliable' on a 2GB card used by the system. Yes you can squeeze in more, but you quickly start playing on the edge of the card dropping out. If you happen to have 2 2GB cards, you can put the system on one and use the other only for Studio and almost double the scene potential.
This post to me is misleading to other people suggesting that everyone will need to buy a $1000 graphics card if they want to render more than 4 people. I humbly request that you try to be a bit more objective with your post so that the DAZ community can continue to embrace Iray and not be scared off by such claims and/or confusing posts like this.
The attached scene has 4 figures with 4 different texture sets, lots of vehicles, air and ground, (although one is repeated), a mec and a HDR background. This scene was around 3.5GB of VRAM. No tricks, just out of the box DAZ products with stock render settings aside from the HDR and dome settings. As best as I can interpret from above, this is not possible, but it is.
OH, and if you are operating on a single 2 GB GPU, you will be happy to learn that your Titan X 12 GB will hold 6 times more figures with unique textures, but will also in reality be closer to holding 10 times more. One disappointment I've had, is these bigger cards seem to need more VRAM just to run themselves. My GTX 980s use about 300mb VRAM just sitting at idle with nothing sent to them. I put the Titan in as my main video card, so don't know what it is at idle. The 660s, it seems like are more in the 150mb range at idle.
I hope this helps.
Sorry not wanting to step in and cause arguments, but I have just been looking for an Nvidia card and came across this http://www.ebuyer.com/707426-asus-gtx-760-directcu-ii-4gb-gddr5-dual-dvi-hdmi-displayport-pci-e-gtx760-dc2oc-4gd5
a 4GB GTX 760 with 1152 Cuda cores
Iray by default is set to compress textures it sends to the video card. If you look on the advance tab of render settings when Iray is the selected render you will see the "Texture Compression" boxes. The first box sets when medium compression starts. The second when high compression starts. Certain textures like normal maps are set to never compress, but the rest use these thresholds compared to their greatest dimension. By default they are set to 512 and 1024. So texture smaller or equal to the medium threshold don't get compress, textures between medium and high get medium compressed and textures above high get highly compress.
If you use the emission shader, I would also recommend going in to canvases, under render advanced doing a second render pass set to emission (you can hide everything but the figure emitting light to speed things)It gives you a render of just the light that you can combine in post. I found it much more controlled than bloom for creating some glow. (just blur it and set it to screen)
here's the final full size render I did http://www.daz3d.com/gallery/#images/71525
and here's the post with just the emission shader at a low value on the left, and that render combined with the light pass on the right http://www.daz3d.com/forums/discussion/56019/P1155/#830360
Are you serious ? I suggest everyone to buy $1000 card? you just said that you rendered 4 figures and some other stuff and it was 3.5 GB ? my GTX 760 card use 150 MB doing "nothing" and that is my main card and I am using win8.1 that dont have Aero effects , my GTX 760 2GB can render 5 full different textured Genesis figures at a moment and 192 non textured figures ..that is my limit
but I don't show 192 textured genesis as a prove that my 2GB can handle it as that is misleading trick , as everyone know the textures will be duplicated and count as 1 and not 192
I think I show more than anybody what you can do on 2GB card and inspired others even NVIDIA was impressed with my gallery doing Interview with me so don't tell me I am scaring off the community from using Iray with my claims
this is ridiculous comment and for that reason I am done talking with you on this subject
have a good day
That explain a lot, thanks Chris
I tested with normal maps and never saw the effect and was wondering why it does not compressed it at all thinking it was feature for the Render Server or something but in this case it explain exactly why I could render 5 fully textured genesis as Iray compressed the textures send to the video card using less video memory ,what is actually awesome feature compared to other programs .. and the mystery is solved
Iray by default is set to compress textures it sends to the video card. If you look on the advance tab of render settings when Iray is the selected render you will see the "Texture Compression" boxes. The first box sets when medium compression starts. The second when high compression starts. Certain textures like normal maps are set to never compress, but the rest use these thresholds compared to their greatest dimension. By default they are set to 512 and 1024. So texture smaller or equal to the medium threshold don't get compress, textures between medium and high get medium compressed and textures above high get highly compress.
BTW new monthly FREE HDRI is out
http://www.aversis.be/hdri/hdri-free-monthly-hdr-map.htm
I didn't make any changes to the beams out the side. I didn't add a shader to them either. I really wanted to have the beam out the front to be kind of emissive but when I turned it into emissive, it would look too solid and wouldn't give off enough light. Instead, I thought about just putting a spot light coming off the bot shining pass the girl. That seemed alright. So I took the beam and gave it a frosty glass shader. I think if I up the opacity and narrow it down to where the spotlight edges touch the beam edges, it will look alot better.
animation render test with a gtx 660 gpu only
render time 61 minutes for 272 frames
without shadows 10 seconds for a frame
music by me as always
https://vimeo.com/129673200
I got really curious... as just two G2F and 2GB of VRAM sounds very strange to me - too much for it.
I really dont understand your argument about the virtual memory a video card should take, a card only takes "virtual memory" if it has shared mem... if not it has the given VRAM and that's all - if that RAM is filled, it is filled and no virtual mem is taken.
Anyway, I just wanted to jump in and show you... 12 Genesis-2-Females, all different textures, even different morphs, some are HD (Iray doesnt care about HD) and the bikini ;) -> picture is attached.
System is still reacts at full speed, no slow down.
The scene just takes a bit more than 3GB - I've attached also the GPU-Z statistics, showing the memory usage of three cards.
@sedor
What do you mean about Iray doesnt care about HD?
My webcomic (currently using Iray) is updated here: http://thefarshoals.webcomic.ws/comics/25
Image is clean, though previous page has gore and past pages have had nudity, fair warning.
I got really curious... as just two G2F and 2GB of VRAM sounds very strange to me - too much for it.
I really dont understand your argument about the virtual memory a video card should take, a card only takes "virtual memory" if it has shared mem... if not it has the given VRAM and that's all - if that RAM is filled, it is filled and no virtual mem is taken.
Anyway, I just wanted to jump in and show you... 12 Genesis-2-Females, all different textures, even different morphs, some are HD (Iray doesnt care about HD) and the bikini ;) -> picture is attached.
System is still reacts at full speed, no slow down.
The scene just takes a bit more than 3GB - I've attached also the GPU-Z statistics, showing the memory usage of three cards.
Yes! That follows with exactly what I'm seeing. 4GB can do a LOT.
Greetings,
It can; I actually treat it pretty differently than everybody I've seen so far. I construct my scene, then do a test render, watching my GPU memory usage. If it's low (2GB-ish) then I up the resolution, add HD to the characters (render subdivide them), look for surfaces that have displacement and increase the tessellation on those, and I tweak the 'fine details' until I get the scene to around 3.5+GB. Then I let it render...But if I've created a complex scene (e.g. using one of Merlin's awesome sets :) ) and I blow out video memory with two characters and a few nice textures, then I back off displacement, angle things so textures are more procedural and less Big A$$ Textures, etc...
Does anyone else do this?
N.b.: I think this is a bug... If I do renders, and cancel them, and do renders, and cancel them, my GPU-Z shows that my video card memory never gets back to zero (it's not being used for system drawing, so it zeroes at around 10MB of baseline usage). And eventually the same scene, without anything extra in it, will no longer render.
Shutting down DAZ Studio and starting it back up again gets me back to a zero video memory state (10MB used). Has anyone else seen this?
-- Morgan