Tutorial How to optimise render times in iray (a small guide)

[recreation]

How to optimise render times in iray (a small guide)



Lately I have seen a lot of questions regarding this topic, so I decided to make this small guide.
While I am by no means a pro and am sure that other people know more and better tricks, I'm going to share what I have learned over the time.
There are a few things that can drastically reduce render times if done right.
The guide is meant for rendering in iray.


​

Contents:

#1. General things.
#2. Importance of lights
#3. How geometry and texture affect the memory consumption
#4. Texture resolution.
#5. Reflectivity
#6. Tools

1. General things.
​

For the people who are completely new to DAZ:
Iray works best with nvida grafics card (cuda). More cuda cores mean faster rendering. Of course Iray CAN use the cpu, but it is way slower, no matter what cpu you have.
If you are rendering on cpu or just don't have the best hardware, this post is especially for you!

Before you start to create a scene, think about what you want to show and what you want to achieve with the scene. Don't just drop a quadrillion assets and figures in the scene, because every asset will add to the render time.
Less is more!
It's always a good idea to think about what you really need for a scene to look right and if you're not happy with the end result, you can always add more stuff later.

2. Importance of lights.
​

If you're using iray to render your scene you will quickly notice that your scenes will end up grainy or noisy. That's because iray needs light. A lot of light. This doesn't mean adding several light sources will make it better. The opposite is the case. Every light source generates shadows, rays and whatnot and needs to be computed, so in the end it will add to the render time if there are too many light sources.
To get good results it's best to use as few as possible light sources. For exterior scenes it's always a good idea to use a hdri light and only add additional lights if needed.
Depending on your scene this might even work in interiors.
More in-depth informations about light can be found further down

You must be registered to see the links

thanks to @Rich for posting over here.

3. How geometry and textures affect the memory consumption.
​

Every object in your scene consists of geometry and in the most cases textures. Geometry must be computed, texture must be computed. The more complex the object/asset/figure is, the more compute power - or in this case memory - is needed and the more time it will take to render. If the needed memory is more than your grafics card can handle, iray will drop to cpu rendering, which - as stated above - is way slower than gpu rendering.

A simple cube without any texture will render way faster then for example a figure like the gen3female with all its diffuse, bump, specular and normal maps/textures. Even if you remove all the textures from the g3f, it won't render as nearly as fast as a simple cube, because of the complexity of the figure.
Keep this in mind when setting up a scene.

It's a good idea to remove objects that are not visible in the scene because it will still be computed.
But what if you have to keep them, for example if they drop shadows that must be visible in the render?
You can still delete the textures from these objects to lower memory consumption/render times. This is extremely important if the objects use large textures.
And this brings us to the next point:

4. Texture resolution.
​

Using 8k textures and above gives you nothing but a lot more memory consumption.
8k textures tend to be tripple the filesize than a 4k texture and up to 20 times bigger than a full hd texture (2mb vs 40mb). Even if you render a 8k image, it doesn't make sense to use 8k textures because most of it wont even be visible. It won't look better than a 4k or even a 2k texture.
Now you might think 40mb is not much, why should I care?
It's simple, because it's actually not just 1 time 40mb. Just a single figure like a gen8 uses atleast 3 textures only for the torso. Now add every other body part, like the arm, legs, or the eyes and you easily go above 1gb with one figure only. And that's without any hair or clothing btw.

Set the texture resolution in relation to your desired render resolution and how much is visible. For example: if you render @ 4k and have a mouse placed somewhere at the edge of the sceen, now let's say the mouse takes up about 10% of the space in the final render image, does it make any sense to use a texture that is 90% bigger than the mouse in the image? No!
I hope this is understandable.

There is one exeption though. If you use hdri maps it's actually better to get higher resolution hdri's then the actual render resolution because they will be stretched around the whole scene.

5. Reflectivity
​

I know, I know, everybody loves shiny things and when in comes to rendering, shiny and reflective objects look great. BUT!
Gloss, reflectivity, refraction and backscattering heavily influence render times. The more shine, the longer it will take to render the scene. A scene with just one simple object, one light and 20 mirrors will most likely take way more time to render than a scene with 100 simple objects and a light.
Keep this in mind when you set up your scene.
If you have objects that are reflective or shiny but dont really need to be, you can reduce the glossyness in the surface tab. Removing glossy maps/textures and subsurface scattering (AoA) also helps to reduce rendering times. Most of the time you can get the same effect without the maps if you play a bit with the slider values.

6. Tools
​

Some helpfull tools for rendering:

You must be registered to see the links

Tool to reduce texture size, remove maps and reduce subdivision levels (easy to use, instructions included)

You must be registered to see the links

Both very usefull if you don't want to wait for your render to start the next one. Save your render, create the next one, save it and render them all over night. There are some other nice tools over in the linked thread too.

Other notable/interesting things/tips​

One thing I read again and again when people ask for advice for getting less grain/noise in their render is that it's a good idea to render in higher resolution and reduce the resolution afterwards. Yes it helps to get rid of some grain/noise and yes a higher resolution doesn't impact the render time that much. But if you have a low end computer you will notice an impact and if you are concerned about render times then it's better to spend some more time to optimise your scene before you hit the render button.

If you have more than one (nvidia) grafics card, every of them will be used for rendering (when activated in the render settings) BUT the overall VRAM wont be stacked. That means if you have two 4gb cards, they wont stack to 8gb. If you have a scene that needs 6gb of VRAM they are basically useless and the rendering process will drop to the cpu.
On the other hand, some people claim that if the VRAM size of the cards differ and the scene is too big for the smallest card, the rendering process will automatically drop to the cpu...
THIS IS NOT TRUE! Iray will use the other card for rendering if there is enough space.
Another thing me and a few others have noticed is that, even when the smaller card does not have enough available memory, and won't be used, the gpu usage can still be at 100%. I wasn't able to get any information about why this happens or what the smaller gpu is computing in this case, but as soon as I find out, I'll post it here.
You can read about GPU usage in more detail in the following post by @Rich

One thing I recently found out is that if the whole scene is bigger than you vram size doesn't neccessarily mean that iray will drop it to the cpu. The scene will first be processed in the system ram and gets converted for rendering. The processed scene and memory consumption will be smaller in the end (That's how I was able to render a 50gb scene on my 14gb vram btw).

If you have several light sources, use optix prime acceleration!
For those who don't know what it is or does: Optix Prime is a low level API for computing light ( especially ray tracing) and developed by nvidia for their cuda cores.
Activating it helps a lot to speed up scenes with different or heavy light.
You can activate it in the render settings under the advanced tab.

On the occassion:
Keep your system clean! Anything that impacts your system, impacts the render times too.
if you have less than 15% of your HDD capacity left, it will impact the performance of both of them drastically
Start deleting unnecessary stuff and you'll get a lot of performance back



TL-DR:
Use nvida (cuda) cards for iray rendering
Use just a few simple but effective lights
Don't use too many objects in the scene
Use simple geometrys
Remove every object/asset that is not visible in the render
Don't use higher resolution textures as nessessary
Highly reflective objects add a lot to the render time


Feedback and other helpfull tips are appreciated and will be added later. I want to make this a nice starting point for beginner and advanced user

 

[Rich]

I posted this in another thread, but @recreation asked me to copy it here. The topic was "how does Daz Studio use multiple GPU cards if you have them installed?" Some of this repeats what was posted above, but...

To summarize:

1. iRay will use multiple GPU's if they are installed, have a reasonably up-to-date drive and are iRay-compatible. (i.e. NVidia, and of a modern-enough make that iRay supports the chipset.) Note that you have to enable both in your render settings.
2. You can mix and match GPU's without any problem - the two cards don't have to have the same specs, nor do they even have to be from the same chipset generation. Basically, if iRay can use the GPU standalone, it can use it in combination with another supported GPU.
3. The entire scene must fit into each GPU for the GPU to participate in rendering a scene. iRay does not "stack" the memory of the two GPU cards. So if you have an 8Gb card and a 4Gb card, a scene that takes 3Gb will render using both cards, but one that uses 6Gb will only use the 8Gb card. A scene that takes 9Gb will drop back to CPU rendering, even though you have a total of 12Gb of VRAM. iRay handles this automatically. (Note - Windows is known to "steal" memory from each GPU, even if you don't have a monitor plugged into it. So an 8Gb card may only have something like 7Gb available for iRay.) It's important to understand that the texture sizes reported by the DS log are misleading - they are the uncompressed sizes, however textures are typically compressed when put into your GPU. Thus, it may look like you're getting 9Gb of textures into an 8Gb card, but that's because the 9Gb reported is the raw size, not the compressed size. This has led to a lot of arguing over memory stacking on the Daz forums. Bottom line - NVidia says it doesn't. Period. At least with the consumer-grade cards that we will all be using.
4. The performance of an individual card is dependent on many items, including the clock speed, number of CUDA cores, chipset, etc. If you have two identical cards, the scene will render in just about half the time as with one. (There's a small amount of overhead in using the second card, but it's quite small.) If the cards have different performance, then the performance gain is more complex to calculate, of course. Remember the old "if John can paint a house in 3 days and Bill can paint one in 4, how long does it take them to paint a house together" math problems? Basically, that's what you're dealing with when you combine two cards of different performance. But it'll be faster than with either card standalone.
5. You typically will get more performance from an overclocked GPU as compared with the same GPU not overclocked, since iRay performance is pretty proportional to clock speed. However, OC'ing a card for iRay may be A Bad Idea unless you have really good cooling. iRay will run your card flat-out for the entire time it's rendering. This is typically a much higher load on the card than you would get using it for gaming, for example. So if you don't have really good cooling, you can fry your card. The first few times you do a serious render with any new card, it pays to monitor the card's temperature to be sure your system can handle it.
6. iRay does not use SLI. The core idea behind SLI is to allow a game to get better video performance by having one GPU render one frame while the other GPU renders the next frame, thus allowing two GPU's to give you essentially twice the video performance of a single one. The SLI connector between the two cards allows the second card to pipe its video over to the card to which the monitor is connected. (And possibly more - I don't use SLI, so I can't speak to it much.) But iRay doesn't work that way - it operates each of the GPU's directly and independently. The folks at Daz recommend disabling SLI when rendering using iRay. I don't know the technical ins and outs of what effect leaving SLI enabled would have.

 

[Rich]

A few additional comments to @recreation great introduction:

1. When discussing "render time," it's useful to remember that your system goes through two phases - the first is spent getting everything set up in the GPU, while the second is actually banging out pixels. Everything that is in your scene (and isn't hidden) goes into the GPU, so that takes time. For a large scene, this can take minutes. All the geometry and all the textures have to be processed. Plus, the textures that are larger than your compression threshold have to be compressed. All of that is done using a single CPU thread - even if you have a 24-core ThreadRipper, mostly Daz will only use one for this. Now, usually the scene loading time is WAY smaller than the actual pixel-banging time, but it's not zero. So not only will removing unneeded textures save you GPU memory, it'll cut down on the setup time required to process it.
2. When it comes to lights, @recreation's comments are correct. But, IMHO, the singularly most important thing about lights is to delete or disable any light that doesn't affect the visible area in your render. iRay "knows about" lights, and understands that they're important. So iRay takes special pains to try to find paths to lights. And if you have a light that's not actually illuminating something that can be seen, all that effort is wasted.
3. If you're looking to reduce GPU VRAM usage, focus on textures. They far and away dominate the VRAM usage. It's not uncommon for textures to be in gigabytes, while the actual geometry (meshes) are rarely out of the megabytes range.
4. One of the big things you can frequently get rid of in renders are bump maps. Bump maps are textures that are used to simulate low-level detail on surfaces. Basically, rather than creating meshes fine enough to handle every wart and pimple, the mesh is created with (effectively) a smooth surface, and then bump maps are used to tell iRay "treat this spot as being a little higher and this spot a little lower." Awesome if you're looking at closeups, but if your camera is more than (the equivalent of) a meter or two from a figure, You Won't See The Difference. Totally removing a bump map saves far more memory than shrinking its size.
5. The same can be said for displacement maps, although they're used less frequently these days.
6. One trick that can significantly reduce render time is to change the Max Path Length setting under Render Settings > Optimization. When iRay us computing the color of a particular pixel, it has to know what light hits it. But that light may be the result of light bouncing off another surface. And that light can be the result of light bouncing off another surface. The default setting is -1, which means that iRay will follow bounces until it decides that there's not enough energy left to worry about. ("-1" equals "unlimited") But for many scenes, setting a value of maybe 4-7 for this can cut down on the amount of computation that iRay has to do. This does change the render somewhat, but the effect is rarely noticeable. When you don't want to do this is when, for example, you have a lot of reflective surfaces, or when you're doing a real close-up on a figure. The eyes in G3 and G8 figures have complex geometry, and get some of their realism precisely because of the way that light reflects around inside.
7. Related to bounces - areas that are directly lit will converge ("de-noise") much faster than areas that are only indirectly lit. So if you have an area that's in shadow, you'll get a lot faster result if there's some light somewhere in the scene that directly illuminates that area, even if only faintly. If every path to an area is indirect (i.e. has to bounce), that portion of the image will be very slow to converge, and will be noisier.
8. Lighting indoor scenes can be a pain. One trick is to remove a ceiling or wall that's not visible to let external HDRI light in. If your scene's walls and ceilings can't be individually enabled and disabled, you can use an iRay Section Node to slice off part of your scene. You do have to configure the node to allow light through - that's a parameter on an iRay Section Node. This can also be useful if you're trying to get a camera into a tight space and a wall is in your way where you want to position the camera - slice the wall off with an iRay Section Node.
9. IRay "understands" regular lights, but doesn't "understand" emissive surfaces. By that, I mean that iRay kind of discovers emissive surfaces accidentally, while its algorithms make specific allowances for lights. Thus, if you have emissive surfaces in your render, it can make the scene slower to converge, even though it seems like there's a lot of light. Basically, iRay handles a "true" spotlight much more efficiently than a "light bulb" made with an emissive sphere.
10. That being said, Ghost Lights (emissive surfaces that are set to a very very VERY small opacity - like 0.0000001) can be very useful in lighting hard-to-light scenes. Essentially, they can make light "appear out of nowhere" in situations where the spotlight itself might be in view. Spotlights have a "Render Emitter" setting - if you turn that off, you won't see the spotlight from the back. But you will see it in any mirror or other reflective surface. Ghost lights won't show up. About the only issue with Ghost Lights is that they aren't directional - they emit light very broadly. This means fewer shadows. Fewer shadows can be a good thing from a render time point of view, but they can make a scene look very artificial, since our eyes are used to there being lots of subtle shadows around.
11. If you need a scene to be dark, Tone Mapping (under render settings) is your friend. Set things up with plenty of light so that iRay is happy, but then reduce the Exposure Value. This will make the final render darker even though there's a fair amount of light. Movies use this trick - they film during the day, but change the exposure to make it look like its night. There's a great tutorial on the Daz forums on how to make a scene look like night while still keeping iRay happy:
    You must be registered to see the links

So there's some of my thoughts on the matter. I realize a lot of it was lighting-related, but lighting makes or breaks a scene, both from iRay's point of view but also visually...

 

[mickydoo]

I have a 1050Ti in mine, here in Australia I can but another for around $350, a 1080 is around $850. I could by another 1050 right now , a 1080 is a still a few months away at best. Any idea of the render time difference between the 2 1050Ti's compared to the 1080?

 

[recreation]

$350 for a 1050? Thats hilarious! I wouldn't buy a 1080 for 850 anymore as well, but thats just me.
But i'd suggest to not invest in a 1050 anymore, especially not for the price.
I got my zotac 1070 amp extreme for 300€ which is roughly $350 on amazon and they ship worldwide.
The jump from my 1060 to the 1070 was the best I could do for my budget and the result when rendering is insane.
I posted this in another thread to give you an example:

Quick test.
I used the same (unoptimised) render and let DAZ do its job for exactly 2min each time:
1060 only: 44 iterations
1070 only: 139 iterations (which is already more than I expected)
both together: 285 iterations

 

[macadam]

its is not necessary to "remove" the assets from the scene you render and that are not visible. you can also "hide" them by simply uncheck the "eye" icon.

just saying, because as it is explained in the OP, it didnt mention that and kind of say "remove" in the complete way.

 

[macadam]

also (but that just a personal technique)... on some render ,mostly very close ones, you can do them in multiples stage and then mix them together.

for exempl, if the render show a char (head and torso) in front of a city. you can render the char, and then render the background city separatly. if you use your light correctly and respect the light direction in both render, when you put the renders back together (photoshop or else, just copy the char in the background) . it will be completly invisible that they have been rendered separately.
(render background as .jpg and char as .png, for exempl, allow transparency on the char.)
you can also do that on inside render, by rendering the scene without outside background (view from the windows for exempl) and then render the outside and copy-paste it on it (with inside scene on top).

yes, that require some post work photoshop,gimp or whatever. but its still wayyyyyyy more faster than rendering the whole scene.
soo if you dont have problem with that, then its a great alternative for some renders

(renders also wont be bigger than others renders, once combined , it will make the same size as a full rendered one)

 

[recreation]

its is not necessary to "remove" the assets from the scene you render and that are not visible. you can also "hide" them by simply uncheck the "eye" icon.

just saying, because as it is explained in the OP, it didnt mention that and kind of say "remove" in the complete way.

This is true of course. When you hide them iray won't compute how the light bounces off of them, but deleting them completely is still the faster way. Hidden objects will still add to the render times because the object itself will still be computed.

 

[macadam]

This is true of course. When you hide them iray won't compute how the light bounces off of them, but deleting them completely is still the faster way. Hidden objects will still add to the render times because the object itself will still be computed.

yeah, but its better if you plan lots of renders, because when its just hidding, you dont need to reload the assets again each time and just activate/deactivate them

 

[jaydeertay]

Does anyone have experience using the eGPU setups?

I currently have the Corsair One Elite with a single 1080Ti. It works great for my still images, however I am wanting to speed things up.

I am guessing iRay would pick up the eGPU? This would be a lot cheaper for me to setup and than the new rig I spec'd out.

 

[SteelyDan14]

also (but that just a personal technique)... on some render ,mostly very close ones, you can do them in multiples stage and then mix them together.

for exempl, if the render show a char (head and torso) in front of a city. you can render the char, and then render the background city separatly. if you use your light correctly and respect the light direction in both render, when you put the renders back together (photoshop or else, just copy the char in the background) . it will be completly invisible that they have been rendered separately.
(render background as .jpg and char as .png, for exempl, allow transparency on the char.)
you can also do that on inside render, by rendering the scene without outside background (view from the windows for exempl) and then render the outside and copy-paste it on it (with inside scene on top).

yes, that require some post work photoshop,gimp or whatever. but its still wayyyyyyy more faster than rendering the whole scene.
soo if you dont have problem with that, then its a great alternative for some renders

(renders also wont be bigger than others renders, once combined , it will make the same size as a full rendered one)

This is what has helped me with rendering. I like it because I can light the characters differently from the background so this helps with interiors. I will often use Ghost lights to light an interior but I found ghost lights make the characters skin look like vampires. This allows me to have nice warm characters regardless of the lighting in the room.

 

[macadam]

This is what has helped me with rendering. I like it because I can light the characters differently from the background so this helps with interiors. I will often use Ghost lights to light an interior but I found ghost lights make the characters skin look like vampires. This allows me to have nice warm characters regardless of the lighting in the room.

i use ghost light 99% of time now. and it end pretty ncie. even dark skinned char look nice.
the only problem with ghost light is that it create a whole "plane" of light , and so, the shadow are not defined perfectly like with any spot light or directional light.

 

[31971207]

Can texture resolution be adjusted within Daz3D like SubD or when people talk about playing with texture resolution they are always referring to outside work in PS or Gimp?

 

[recreation]

Can texture resolution be adjusted within Daz3D like SubD or when people talk about playing with texture resolution they are always referring to outside work in PS or Gimp?

Natively no, but there is an addon called "Scene Optimizer" which let's you reduce the resolution of the textures inside of Daz.
You should be able to find it in here in the Asset section.

 

[macadam]

okay.. just little tip.

turn on the post denoiser like in that pic


for "post denoiser start iteration" :

if you have a powerful pc, just put the limit at around 500 or 1000 (depend of scene)

if you have a shitty pc, put the limit around 150-200 (depend of scene, for very big scene, go up to 700, for small scene you can go down to 50)


this, can allow you to finish great renders wayyyyy more faster by denoising the render directly while its rendering, wich grant smooth renders way faster and make you win time very fast.

 

[recreation]

okay.. just little tip.

turn on the post denoiser like in that pic

View attachment 686060
for "post denoiser start iteration" :

if you have a powerful pc, just put the limit at around 500 or 1000 (depend of scene)

if you have a shitty pc, put the limit around 150-200 (depend of scene, for very big scene, go up to 700, for small scene you can go down to 50)


this, can allow you to finish great renders wayyyyy more faster by denoising the render directly while its rendering, wich grant smooth renders way faster and make you win time very fast.

Okay, a little tip from my side regarding the denoiser: Don't use it if you don't have to!
It doesn't matter how long you render or when the denoiser sets in, it WILL be visible.
If you have to use it: it doesn't matter when it sets in, it matters how long the render runs, you can set the denoizer start to as little as 1 iteration and let it render 1k iterations, it will look the same as if you set it to 999 iterations and let it render 1k iterations.
Most important thing to remember is that the more iterations the render has, the less muddy and blurry the image will look with the denoizer.

 

[macadam]

Okay, a little tip from my side regarding the denoiser: Don't use it if you don't have to!
It doesn't matter how long you render or when the denoiser sets in, it WILL be visible.
If you have to use it: it doesn't matter when it sets in, it matters how long the render runs, you can set the denoizer start to as little as 1 iteration and let it render 1k iterations, it will look the same as if you set it to 999 iterations and let it render 1k iterations.
Most important thing to remember is that the more iterations the render has, the less muddy and blurry the image will look with the denoizer.

i know.. i just mentionned that for people who may have bad computer and so cant let render run forever....

i know how it feel to try doing a simple render and wait for like 10 hours for render 1 char alone :/ and ending up with finally a render that is very grainy.

this is still the best option for helping with low setup.

 

[recreation]

i know.. i just mentionned that for people who may have bad computer and so cant let render run forever....

i know how it feel to try doing a simple render and wait for like 10 hours for render 1 char alone :/ and ending up with finally a render that is very grainy.

this is still the best option for helping with low setup.

I know that feeling too, I did my first renders on CPU and it took months until I was able to buy a cheap gtx 1060 xD
But still, I think it's important to know that the denoizer should only be used when really necessary.

 

[macadam]

I know that feeling too, I did my first renders on CPU and it took months until I was able to buy a cheap gtx 1060 xD
But still, I think it's important to know that the denoizer should only be used when really necessary.

made a quick test...

this is a 200 iteration stopped + denoiser.. 0% rendered and less than 4 minutes.

Spoiler

You don't have permission to view the spoiler content. Log in or register now.

soo yeah, its not perfect for sure, a bit blurry here and there, a bit of bad stuff.
but i bet a 3 minutes on photoshop and that came out pretty nice. (at least for me)
and honestly, even without photoshop, this is still far better than some grainy bad renders i saw in some games around...

soo yeah, that post denoiser things IS VERY USEFULL for sure .

(sure its 720p and all, but like i mentionned, for who have powerful pc, just turn up the start iteration. for who have low pc, this is perfect imho)

 

[galora]

made a quick test...

this is a 200 iteration stopped + denoiser.. 0% rendered and less than 4 minutes.

Spoiler

You don't have permission to view the spoiler content. Log in or register now.

soo yeah, its not perfect for sure, a bit blurry here and there, a bit of bad stuff.
but i bet a 3 minutes on photoshop and that came out pretty nice. (at least for me)
and honestly, even without photoshop, this is still far better than some grainy bad renders i saw in some games around...

soo yeah, that post denoiser things IS VERY USEFULL for sure .

(sure its 720p and all, but like i mentionned, for who have powerful pc, just turn up the start iteration. for who have low pc, this is perfect imho)

I can't say I agree, you sure aren't wrong I can't say it since I don't know.

What I want to say and I guess you'll know already but there are the denoisers from nvidia and intel (I seen to remember recreation saying the intel one was better?)...

I'm testing them out and they are amazing, at least compared to the post denoiser.
GUI by CulayTL