Render Speed, physics comparison

Honzo

I have only a little experience with either Blender or Daz, but it seems to me that Blender is much faster, and Daz is a little easier to use for us noobs. For instance, the IRAY rendering seems to be quite slow, even for a very low resolution image, and dForce physics take about ten minutes with dForce hair- for a single pose. I'm running a Ryzen 7 with an Nvidia RTX 2060 Super, so I don't expect miracles, but in my very limited experience that's about ten times what Blender would take on the same hardware. Is this inherent in Daz, or am I doing something wrong?  I've tried to follow a couple of the site's tutorials, but I still get the slows.

Perhaps the best question is: what is the fastest way to apply physics to hair? 

Padone

You can use diffeo to import daz figures in blender and convert hair to particles. Then I'm trying to get Thomas into better dforce support, but I'm not sure if he agrees. So for the time being you have to setup the simulation yourself.

As for speed, in my experience cycles is overall faster because of the better denoiser, but as for "brute force" rendering iray is faster. As for simulations, again dforce is faster but blender is more stable, that means it doesn't explode so often. Then of course blender gets much more simulation tools, while dforce is limited to cloth and hair so far.

http://diffeomorphic.blogspot.com/

https://bitbucket.org/Diffeomorphic/import_daz/issues/410/better-dforce

 

 

tapanojum

I have to disagree with both assesments of Iray or Dforce being faster than their Blender counterparts. Perhaps there is some truth to this when measuring Cycles performance when using the default Diffeomorphic conversions which tries to replicate Daz. For example, the skin shader looks nice and does a great job replicating how the character would look in Daz Iray but it's incredibly slow. Hair mesh to particle conversion looks nice but again, using the default values is inefficient. There are far too many parent guides and unless you manually go in to heavily trim this count, simulation physics will be slow. Blender cloth physics is a joke compared to Marvelous Designer but still performs much faster and  more stable than Dforce in my experience.

 

TheKD

In my iray vs cycles tests, speedwise without leveraging the cycles denoiser goes to iray, but not by a whole lot. But when using the optix denoiser, cycles blows iray out of the water in my tests. These are in my more production level testing, rendering large images. I can render a lot larger in cycles without maxing my vram, one of the biggest pulls of cycles.

benniewoodell

tapanojum said:

I have to disagree with both assesments of Iray or Dforce being faster than their Blender counterparts. Perhaps there is some truth to this when measuring Cycles performance when using the default Diffeomorphic conversions which tries to replicate Daz. For example, the skin shader looks nice and does a great job replicating how the character would look in Daz Iray but it's incredibly slow. Hair mesh to particle conversion looks nice but again, using the default values is inefficient. There are far too many parent guides and unless you manually go in to heavily trim this count, simulation physics will be slow. Blender cloth physics is a joke compared to Marvelous Designer but still performs much faster and  more stable than Dforce in my experience.

@tapanojum I think it depends on your graphics card for iray being faster. I have a 3090 and in iray if I turn on post denoiser, I can get a render at 1920x1080 (I do animations) of a G8 character in 30 seconds that does not have that kind of a painted denoised look only needing about 350-500 iterations, but that exact same shot I'll port into Blender with diffeomorphic and render in cycles at 256 samples, and even with the open image denoise turned on (which I think is way more amazing than post denoiser in Daz) it still takes on average 2 minutes to render the exact same shot. I go back and forth between the two programs on if a specific shot would look better in Blender because sometimes the lighting I'll have set up will really make the skin pop and lays over it so beautifully, and other times it looks better in Daz, so I always set up the shot in both and if it's better for Blender I'll gladly wait for the two to three minute frame renders. The good thing is the materials in Diffeomorphic are at a place where I can mix Blender and Daz shots and you'd never really know unless you know exactly what to look for, but for a casual viewer, they would never notice it. 

@Padone, that would be amazing to get more dForce support, I hope you can convince Thomas to go further with this. 

Honzo

Thanks for all the feedback, everyone. I suspect that I'm setting things up wrong in Daz for both physics and iray. There are a lot of little details to get down, and I think I have some confused memories from a year ago when I tried to do a little of this, so I really need to back up and start over again from scratch.

marble

Padone said:

You can use diffeo to import daz figures in blender and convert hair to particles. Then I'm trying to get Thomas into better dforce support, but I'm not sure if he agrees. So for the time being you have to setup the simulation yourself.

As for speed, in my experience cycles is overall faster because of the better denoiser, but as for "brute force" rendering iray is faster. As for simulations, again dforce is faster but blender is more stable, that means it doesn't explode so often. Then of course blender gets much more simulation tools, while dforce is limited to cloth and hair so far.

http://diffeomorphic.blogspot.com/

https://bitbucket.org/Diffeomorphic/import_daz/issues/410/better-dforce

 

 

Interesting post that you linked there - the "Better dForce" one. Something caught my attention and surprised me though ...

The dforce time on my system is 45 sec, so 100 sec for blender seems acceptable considering that dforce is using the gpu while blender is using the cpu. Also I have a weak cpu ryzen 3 2200G, that’s slow compared to my GTX 1060. 

 I thought the same until I upgraded my GPU from a 1070 to a 3090. I have not seen any improvement in dForce simulation times at all since the upgrade. Obviously my render times are significantly better but not dForce.

Any idea why?

 

tapanojum

benniewoodell said:

tapanojum said:

I have to disagree with both assesments of Iray or Dforce being faster than their Blender counterparts. Perhaps there is some truth to this when measuring Cycles performance when using the default Diffeomorphic conversions which tries to replicate Daz. For example, the skin shader looks nice and does a great job replicating how the character would look in Daz Iray but it's incredibly slow. Hair mesh to particle conversion looks nice but again, using the default values is inefficient. There are far too many parent guides and unless you manually go in to heavily trim this count, simulation physics will be slow. Blender cloth physics is a joke compared to Marvelous Designer but still performs much faster and  more stable than Dforce in my experience.

@tapanojum I think it depends on your graphics card for iray being faster. I have a 3090 and in iray if I turn on post denoiser, I can get a render at 1920x1080 (I do animations) of a G8 character in 30 seconds that does not have that kind of a painted denoised look only needing about 350-500 iterations, but that exact same shot I'll port into Blender with diffeomorphic and render in cycles at 256 samples, and even with the open image denoise turned on (which I think is way more amazing than post denoiser in Daz) it still takes on average 2 minutes to render the exact same shot. I go back and forth between the two programs on if a specific shot would look better in Blender because sometimes the lighting I'll have set up will really make the skin pop and lays over it so beautifully, and other times it looks better in Daz, so I always set up the shot in both and if it's better for Blender I'll gladly wait for the two to three minute frame renders. The good thing is the materials in Diffeomorphic are at a place where I can mix Blender and Daz shots and you'd never really know unless you know exactly what to look for, but for a casual viewer, they would never notice it. 

@Padone, that would be amazing to get more dForce support, I hope you can convince Thomas to go further with this. 

Most of Daz stuff is non-pbr and built specifically to look good in Daz Iray. Diffeo does a good job at converting these materials to look like Iray but this usually comes at the cost of rendering speed. I just wanted to point out that it's not accurate to compare cycle speeds to Iray when using default Diffeo conversion. Of course within the context of this being a Daz forum, this may be a practical comparison for people simply wanting to use Daz inside of Blender with as little hassle as possible. For others who may want to get more involved and eventually ditch the default Diffeo conversion, cycles should provide faster speeds unless there's been some drastic changes since I last tested the two engines 6 months ago using simple shaders.

tapanojum

benniewoodell said:

tapanojum said:

I have to disagree with both assesments of Iray or Dforce being faster than their Blender counterparts. Perhaps there is some truth to this when measuring Cycles performance when using the default Diffeomorphic conversions which tries to replicate Daz. For example, the skin shader looks nice and does a great job replicating how the character would look in Daz Iray but it's incredibly slow. Hair mesh to particle conversion looks nice but again, using the default values is inefficient. There are far too many parent guides and unless you manually go in to heavily trim this count, simulation physics will be slow. Blender cloth physics is a joke compared to Marvelous Designer but still performs much faster and  more stable than Dforce in my experience.

@tapanojum I think it depends on your graphics card for iray being faster. I have a 3090 and in iray if I turn on post denoiser, I can get a render at 1920x1080 (I do animations) of a G8 character in 30 seconds that does not have that kind of a painted denoised look only needing about 350-500 iterations, but that exact same shot I'll port into Blender with diffeomorphic and render in cycles at 256 samples, and even with the open image denoise turned on (which I think is way more amazing than post denoiser in Daz) it still takes on average 2 minutes to render the exact same shot. I go back and forth between the two programs on if a specific shot would look better in Blender because sometimes the lighting I'll have set up will really make the skin pop and lays over it so beautifully, and other times it looks better in Daz, so I always set up the shot in both and if it's better for Blender I'll gladly wait for the two to three minute frame renders. The good thing is the materials in Diffeomorphic are at a place where I can mix Blender and Daz shots and you'd never really know unless you know exactly what to look for, but for a casual viewer, they would never notice it. 

@Padone, that would be amazing to get more dForce support, I hope you can convince Thomas to go further with this. 

Most of Daz stuff is non-pbr and built specifically to look good in Daz Iray. Diffeo does a good job at converting these materials to look like Iray but this usually comes at the cost of rendering speed. I just wanted to point out that it's not accurate to compare cycle speeds to Iray when using default Diffeo conversion. Of course within the context of this being a Daz forum, this may be a practical comparison for people simply wanting to use Daz inside of Blender with as little hassle as possible. For others who may want to get more involved and eventually ditch the default Diffeo conversion, cycles should provide faster speeds unless there's been some drastic changes since I last tested the two engines 6 months ago using simple shaders.

Padone

@tapanojum Of course cycles can be faster if you use principled instead of the volumetric skin. But so can be iray if you use simpler shaders. To compare the speed of two rendering engines we need to use similar shaders, if we want to be fair. Then diffeo does provide a principled option for materials so you can get the better speed. Also please read my dforce conversion attempts below, if you can help to speed up the blender simulation it's great. In my tests dforce is about twice faster, but I have a weak cpu so I guess it's expected.

https://bitbucket.org/Diffeomorphic/import_daz/issues/410/better-dforce

@marble You can see that dforce is using the gpu in the task manager, by selecting the cuda graph. Then why a 3090 isn't faster than a 1070 I have no idea, it should be. You may check the cap in gpu-z to see if there's something.

@benniewoodell Yep me too hope to get Thomas into better dforce, we'll see ..

wolf359

@tapanojum

You are correct!!
Volume based SSS is retarded unless you are trying to exactly replicate the look
and performance of the  brute force path tracing in Iray. 
which is what Diffeo conversion does

I do understand this Daz site "Blender" forum is really just a place to discuss
Blender still image rendering as an alternative to slavery to NVIDIA.
and that liberation from one GPU seller is good for us all.

However comparisions between render engines is not technically valid unless you compare ALL features.

if you compare the animated filmaking features of cycles
to Iray... it is no contest...none.

Go ahead, invest in a pair of 3090's or a render farm or whatever

You will never acheive a truly cinematic film look without MOTION BLUR, DEPTH OF FEILD/BOKEH
and the  VOLUMETRIC LIGHTING EFFECTs that IRay does not offer IIRC.

 

marble

Padone said:

@marble You can see that dforce is using the gpu in the task manager, by selecting the cuda graph. Then why a 3090 isn't faster than a 1070 I have no idea, it should be. You may check the cap in gpu-z to see if there's something.

 

 I did check GPU-Z and I do see the GPU in use (between 50% and 70% as I watched). So it is odd that the time taken to simulate a garment does not seem to me to be any faster than it was with my old GPU. I also checked the dForce settings to make sure my 3090 is selected.

Padone

wolf359 said:

Volume based SSS is retarded unless you are trying to exactly replicate the look and performance of the  brute force path tracing in Iray.  which is what Diffeo conversion does

To avoid misunderstandings, diffeo does have a principled option for materials, that doesn't use the slow volumetric skin. But then the material conversion will be approximated because the principled shader can't do everything iray does. So it's the user choice speed vs accuracy.