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
That's very interesting. Well I was referencing the benchmarks on the official Auto Size Addon page linked above. If there is that difference then of course it's worth a change in tile size. Given that the denoiser stills working fine I mean.
Where can I find the benchmarks you're referring to, can you link please ? Or are they tests you did on your pc ?
EDIT. Never mind I did see your post above and going to check though I'm not a big fan of Blender Guru. We cross referenced. You're writing too fast or I'm too slow
I'm not sure what point you're trying to make. Yeah, you can change a ton of settings in your render to make it much faster. But I thought you had mentioned that a certain tile size works for both GPU and CPU. Clearly, that's not the case if you're talking about apples-to-apples comparison of exact same results.
In any case, for those who care there is an addon in Blender that automatically chooses the correct tile size for your hardware and the scene to optimize it for fastest results. I've been using it for a long time, and it does a great job. Highly recommended.
Not really; not even close.
They are different, and have strengths and weaknesses.
But think how many cores the GPU needs in comparrison to the 16 threads on your CPU; it isn't performing hundreds of times quicker, but on the 1080ti ten times quicker for a lot more than ten times the cores.
Then, without a CPU, it would never get to the GPU, and when the GPU hasn't the RAM it falls to CPU.
GPUs aren't a nirvana, which sounds more like hype from Nvidia's marketting guys.
The point I'm making is that the denoiser is sensible to the tile size too. If it's too large then it may go out of memory. And the denoiser is your best friend for speeding up the rendering. So when you choose a tile size you have to get the overall best. In my post above I was indeed referencing the benchmarks in the Auto Size addon page you're advising. I also see that the Blender Guru video is before 2.79 when the denoiser was introduced.
Keep in mind that the denoiser is nice, but doesn't give you something for nothing. I view it as more along the lines of a post-processing blur effect that blurs the noise, rather than actually caculating the real values to remove the noise. And depending on what you're rendering, you have to do a lot of tweaking of the de-noiser to make the noise look better, and in some cases it may not make much difference in overall render time to get desired results.
If there was a magical way to click a check mark to get the same results in a fraction of the time, then it would be part of the Cycles (or other) renderer, not as a special effect that you can play with.
Not even close. The denoiser in Cycles is not a post-process effect. That's why it's so good. It gets the actual information from the rendering pipeline itself. And it does actually "caculate the real values to remove the noise". Compared to an external denoiser such as GIMP or Photoshop there's no match at all. And personally I found the default values are quite good. I tweak a lot in the integrator to make it faster depending on the scene I have to render. But I never had to tweak the denoiser so far.
Is that right? I recall seeing a video a while back where the guy did some comparisons using a wine glass (?) or something like that with complex reflections/refractions, and he had to tweak the de-noiser settings a lot, and the result was that the noise became blurred unless he cranked the samples way up. I guess, as with anything in the rendering world, it depends.
But again, I'm not sure how the de-noiser relates to our discussion of GPU threads, other than your concern about additional VRAM, which seems to be a bit of a side issue?
A few years ago I had a conversation with a senior executive at Pixar who commented that the company was always interested in hiring artists who were trained in game development techniques because they were incredibly efficient with their modeling and rendering practices. Games have to render in frames per second, not frames per hour, and that requires significant discipline and technique for all those polygons, shaders, and textures to rip through the engine as fast as they need to. Hollywood rendering takes a great deal of time per frame, even after the tech artists get a crack at them, but instead of asking their artists to take the time to be more efficient they are solving the problem with larger and larger render farms.
The bottom line here is that there could be some significant optimizations you could make in your scene before you need to start throwing money at your render solution.
Okay, I found the video I saw that discusses de-noising. I pretty much dismissed it at the time I saw it, mainly because the de-noising checkbox was in the Render Layers section, so I assumed (maybe mistakenly) it was more of a post effect. I'm certainly not an expert whatsoever on de-noising, but I woudn't be surprised if it is an effect that is applied to the image after the scene is rendered. And since any post processing, by definition, uses info from the render pipeline, I guess the devil is in the details of how it actually works. Either it sends rays from the camera to the scene and calculates bounces (which takes the time), or it does something after all that hard work is done.
https://www.youtube.com/watch?v=2yhqIyagDQM
Cycles has problems with caustics if that was the test. Also in animation there are custom shaders for glass since a "real" rendering would be too slow anyway. So complex refractions are not really a good test for animation in my opinion. That's more for still pictures of architectural rendering. And vram is the one you can't expand but changing the card so it does always matter.
Wow, that's interesting. I assumed that they'd have a strong push to make everything more efficient because that affects the bottom line. But I suppose that if you focus on quality, and hire artists with artistic skills and talents, you may not be getting those who are also efficiency and optimizing experts. Interesting. I guess it makes sense they'd now focus on game developers to get the best of both worlds.
Interesting...out of curiosity I went into Blender and loaded a random scene, and rendered it with and without de-noising checked. No changes, just checking and un-checking de-noising.
Without de-noising it took 35 seconds, and with it took 40 seconds. And I noticed that, with de-noising enabled, after a tile is rendered in the preview it will erase the tile and re-render it. But it doesn't do that with de-noising off.
So it makes me think that maybe it is some sort of post effect.
It means your render settings are "too good" already and don't need a denoiser. To get the taste of the real "denoise power"
do set your samples down to say 32. Then the rendering will suck. And then turn on "the magic". Below an example with the G2 starter pack. With and without denoise.
And back to iRay, there's this interesting discussion where the guy renders his animation in about 10-20 seconds per frame. I believe he's doing the denoise in post with After Effects since it seems to me a too good time for iRay alone. No matter the rig.
https://www.daz3d.com/forums/discussion/87211/daz-studio-nvidia-iray-interactive-photoreal-animation-tests/p1
Ahh, okay. So it's more about cranking down the samples to shorten render times, but at the same time the de-noiser will improve results. I thought it would decrease render time automatically based on achieving the desired convergence or whatever using de-noising.
Which makes me even more convinced it's a post processing effect.
But still that's pretty cool how it cleans up the noise in such a bad render result. So do you do multiple renders with different samples, and tweak the samples until the result is where you want? Which kinda defeats some of the purpose since you need to re-render until the de-noise gets it right?
Of course not. Animation is not the same as still pictures. You can't let each frame go on till convergence no matter what. You have to compromise to get speed.
Usually when you do a short there are multiple scenes. Each one has its materials and lighting conditions. So I do a few frame tests first with the integrator to get a rendering that's good enough without the denoiser. Then I switch it on and render the whole animation. That's how it's supposed to be used I guess. But if you find a better way you're welcome.
And it's not a post. In Blender post-filters have their place and are applied after rendering. You can see the change in real time. That's not how the denoiser works. It's tied to the rendering pipeline.
Wow, thanks Padone. I got a 20 second render that looks FAR better than a 40 second render
I'm officially a de-noiser convert.
When you are looking at a still image you have time to examine it closely and see all the subtle detail. In an animation the movement can distract you from seeing imperfections. If you freeze frame most DVD movies you can see that the picture quality is not as good as it looks when it is running especially when there is a lot of movement.
One approach would be to run short test renders with the quality settings reduced and find one that gives a reasonable balance of quality vs render time.
@ebergerly Welcome to the "elite club"
.
@Peter Wade That's exactly what I mean.
In animation, the more time you can shave off rendering each frame, the better. You can't let the render go on forever. So it's a compromise, and with noise reduction or using NeatVideo you can save a ton of time without the finished product looking bad at all. Most of the time users spend doing renders here is to reduce noise. A de-noiser is what you want.
Okay, so now we need to complain that DAZ Studio doesn't have a de-noiser feature like Blender
Very cool. Here's a 9 second render with only like 100 samples and de-noising on. And the same render with de-noising off and 256 samples is all noisy and ugly.
The very fact that you actually have these intelligent in render
optimizations in Blender( and other engines), is the reason I so often refer to
Daz Iray as a "Dumb" Brute force path tracer.
Have you looked at the options under Branched path tracing
in blender?
Much render time can be saved by optimizing the number
of samples on a per surface type basis.
This allows you to decide where the time saving compromises
will be implemented( glass,metals etc)
without you being forced to make Global
optimizations that affects the entire image.
This is particularly important to a sci fi animator like myself who use alot
of reflective metals and such.
Daz Iray is quite lacking in the ares of optimization and too dependant on
the Nvdia hardware to which it is attached.
it was not actually me who suggested the denoiser I use blur and motion blur in virtualdub
I also use a few filters in Hitfilm express
there is also Mcasual's despectler but you need to name your image files and have a corresponding text file to batch convert image series, I have never done it only individual ones.
...
I'm wondering if anyone in this thread knows what Optix is? If so, they might know what the upcoming newly released Nvidia Optix 5.0 is.
I wonder if it will be like motion blur which hasn't been enabled for us after a couple years. 3Delight has motion blur but Iray doesn't.
3Delight also has micro-displacement that Iray doesn't. That's why you have to crank-up subdivision till death with the HD morphs. As I said, I'd never use DAZ Studio for animation unless I'm forced.
As for me, even if the denoiser does a great job in Cycles, it's already old stuff. I'm waiting for EEVEE coming out this year for real-time PBR. That's the real next revolution in animation. Meanwhile iClone 7 is already doing a good job.
I think that Optix Prime acceleration used to be an option for Iray renders but I can't find it in the render settings for Studio 4.10. Has it been removed, hidden somewhere or did I just imagine it?
From a short web search I think that Optix may be the lower level engine that Iray is built on, but I may have got that wrong. If I'm right, version 5.0 probably means they have upgraded the whole thing a bit.
Optix 5 is a denoiser. Something to use Titan V tensor cores on. Not sure if Daz Studio will get the AI.
We've always had the ability to use OptiX acceleration for Iray in DS. it's found in the Advanced tab under render settings (see below).
From here:
From here we get a bit more about OptiX 5.0. Iray in DS currently has a noise filter, it seems the difference with OptiX 5 is that is has an AI-Accelerated Denoiser .... from the article, I certainly hope we get it in DS!!!
Improvements in OptiX 5.0
...and who here can really afford to drop 3,000$ on a GPU card?