Blender Cycles, Renderman & Redshift Render Comparison

hello michael here with another renderer comparison after the last video there was a few people that asked me to compare renderman in comparison to the other renderers so today we're going to have a look at this scene here rendered in both renderman redshift and cycles because the scene was originally designed to be rendered with renderman what we're going to be doing here is using that as our baseline render and then trying to get the quality of all the other renderers to the same level and while i've tried my best to get the scenes to look relatively the same in each you will notice some minor differences in things like color and light value however in each scene i've used the same amount of lights and tried to keep the workflow and settings as similar as possible and hey make sure you're subscribed with notifications on otherwise you may be missing out on the many tutorials that we're releasing for free each week here on youtube so here is just a quick look at this scene inside maya the main thing to point out here is that i am using an environment light that is linked to this portal light so all this environment light is doing is being fed through this one particular region here aside from that all i've got set up in the way of lights is some extra lights where i wanted some more illumination emulating bounce light from the surface of this desk or in this area here where i wanted a little bit more illumination on the side of the amplifier we also have the vu meter here which i have got some green lights firing out of and then some extra ones to illuminate the area around i didn't use spherical lights where some people might have or point lights if you're coming from blender because of the inconsistency with control and the extra noise it can introduce by having multitudes of light rays being cast in all directions not every material is set up with full pbr textures however a lot of them will include specular maps diffuse maps bump maps and roughness maps as well i'm not going to go through every single setting that i used here but the main thing to point out is that i'm using a max number of eight bounces and then the displayed sample rate for each render here renderman will have a distinct disadvantage time-wise compared to the other renderers because i am using an older cpu compared to my much newer gpu also with renderman i'll just point out that it is the renderer that i am most experienced with for the last couple years i have been making tutorials for it and i'm also featured on the renderman website so if you want to include that as bias feel free to however that's just the way it is okay so let's have a look at some raw renders straight out of renderman these have all been denoised using the renderman proprietary denoiser which occurs after the batch render is complete so this image on the face of it doesn't look too bad this is at 16 samples however you'll see it starts to fall apart in some of these shadowy areas around the corner of the room and it gets a little bit chunky under the desk there and to the left hand side of the amp as well so we're going to increase the sample rate to 32 and see how much cleaner there is so you can see it's improved a little bit but we're still experiencing noise in the same areas also you'll notice that the render time has essentially doubled with render man basically as you double your samples you will double your render time so you can pretty much set your watch by it so let's increase the sample rate again to 64 and see how much we improve okay you can see now at 64 samples that we're getting very close the synthesizer board is pretty much spotless at this point and the corner of the room has cleaned up a lot though it is still a little bit dirty that lower left hand side also a little bit dirty and under the desk again it's still a little bit noisy however if you are paying attention to the fully composited shots from the beginning of the video you'll notice that a lot of these areas are actually out of focus and i knew this going in i'd plan this shot so the further away areas i could put out a focus so i wouldn't have to translate them to a completely clean render unfortunately though it looks pretty good as a still it doesn't really work very well as an animation the noise becomes too apparent so let's increase it to 128 which is actually the final sample rate that i decided to go with balancing the amount of time that i had to complete this render and the quality that i was looking for and if we zoom in on these trouble spots that we had previously like the corner of the room here you'll see that the noise is all but gone it is not 100 perfect but as i stated it is going to be out of focus so it doesn't matter as much and the area next to the amplifier is much cleaner as is the occluded shadow area underneath the disk there and near the floor the foreground elements which are going to be in focus are perfect so we're really happy with that and we're getting all that nice detail off our incredibly fake borg synthesizer so let's compare our render times between the four different sample rates we chose to have a look at there and you'll see that the render time pretty consistently increases as we double the sample rate the total render time at the target quality which was 128 samples came out to be 28.3 hours so i rendered this over three days well actually three nights as it were and you will have also noticed that i had a couple of other shots at the beginning there so they also attributed to probably about a week and a half worth of rendering overnight so that's why we're obviously trying to plan out our shots to be able to minimize the amount of rendering time we have to use where we can use the denoiser as much as possible and then fix up the rest and post with some defocus and some other little tricks that i used next we're having a look at redshift you'll see the render settings there on the left but again the most important ones that we're going to look at here are obviously the sample rate which will be displayed on screen per render and also that the gi bounces are set to 3 and we're also using irradiance point cloud as our secondary gi engine and that is set to one sample per pixel i was just going to use the single engine with redshift but i found that the renders came up a little bit dull looking though they were rendering faster i thought it best to just try and get a better result rather than the fastest time the other thing you might be realizing is that the renderman version used eight indirect bounces so we're using less indirect bounces with redshift and i'm going to do the same with cycles the main reason for this is because i didn't want to spend the rest of my life rendering so i decided to reduce the amount of gi bounces for the two render engines that are going to be using my gpu rather than my cpu so you can see that comparison a little bit more directly and as the 16 sample version of the register scene comes into view i'll just point out that i have been using redshift for five years or thereabouts since one point something rather and it is currently in 3.0.34 or at least that's the version i'm currently using so let's have a look at the scene here you can see that i only took 14.56 seconds to render up and that's pretty fast however we do have some issues in the scene you'll see that in the top middle corner we've got some noise there from the optics denoiser which was enabled for this one and the wallpaper in general on that wall is having a little bit of issues optics isn't quite sure what to do with it and it's kind of made it look a little bit wobbly some of the other darker areas like under the desk there has got some artifacting which isn't the best but otherwise the scene has rendered up reasonably well at 32 samples we get a render time of 20.58 seconds and you'll see that optics is starting to figure out what is going on with this 70s wallpaper though it's still not quite sure we're still seeing some wobbling there as well as a little bit of artifacting up in the corner there near the shelves as well as this area next to the back leg of the desk near the wall this highlight is actually starting to get a little bit more artifacted it would seem seems like optics doesn't really know what to do with this particular area at 64 samples obviously we see an improvement again and our render time increases to 30.19 seconds the wallpaper is starting to clean up nicely though it's not quite there yet as well as those occluded shadow areas and near the shelves that area under the desk is still seeing quite a bit of artifacting and it seems that optics is having some real trouble with it moving on to 128 samples at a time of 42.69 seconds we're getting very very close to what i would consider production ready we are still experiencing some issues with that area under the desk there but knowing what we know from the renderman version that we're going to be putting that out of focus it may not be too much of an issue one other area that i will point out which actually uses the same material is the window cell it does have a some of that similar artifacting so i wonder if it is something to do with that specific material and the lights interacting with it that it doesn't like that's probably something worth exploring in a different video though now rather than optimizing i decided i would just brute force this i did optimize in the previous video and a lot of people thought that was unfair so i'm just going to increase this sample rate which is not the best way to work but i'm going to just show you for the sake of showing you so here is redshift at 256 samples and you'll see that the render time increases to 68 seconds so it's starting to get a little bit longer there but it is pretty much cleaning up those two areas that we were having issues with being the window cell and that area under the desk a very small amount of artifacting though but at this sample rate and this time we're still looking at a time and a quality or cleanness that is much better than the renderman version so because it's relatively quick to render a single frame i just increased the sample count again to 512 samples and we got a render time of 94 seconds and you'll see that our returns are diminishing the difference between this version and the 256 and even possibly the 128 sample version considering we are going to be doing a bit of compositing in the end to this is possibly getting negligible so if you are looking at saving time or you're running a lot of frames then you may not need to sample to this higher sample rate so obviously the most important thing with an animation is how it animates and you can see at 128 samples we're not looking too bad in fact we are actually getting a lot less noise than we got with the renderman version so that is good the artifacting that we were seeing under the table there and on the windowsill are still present at the sample rate however and you'll also notice a small amount of flickering just under the keyboard there so if you are not compositing and you don't have a depth of field that is going to save you here then obviously you would have to do something about that i wouldn't necessarily recommend what i'm doing here which is just brute forcing it and increasing the sample rate but that's what i'm going to do so let's have a look at what it looks like at 512 samples so you'll see the flickering under the table and on the windowsill there are reduced again at the sample rate and i'd also like to point out i did use film grain in the final composite so you would probably not be able to see any of this at all at this rate and considering that the render time is only 86 seconds per frame at the sample rate you could probably get away with brute forcing it though i would probably want to go in and figure out why we're seeing that flickering is it something that we can clamp out is it something that we can adjust with a material is it something that we could actually mask and deal with in post very simply these are all options that i would explore before just brute forcing it and learning nothing so here is our reference renderman animation at 128 samples this is the raw version the non-composited version and here is the redshift version at 128 samples and you can see that you'd be quite happy with the redshift version at this rate the noise is actually significantly less so we know that with that threshold if we add the compositing in you can barely notice any of this noise that may be occurring in the raw non-composited version and in case you were curious i decided to render out the 512 sample one as an animation and you can see that while it may be a lot cleaner the gains that you were getting are probably negligible as i mentioned earlier though if you've got the time it is always best to go for the little bit of extra quality if you can and finally let's move into blender you can see here it being rendered in the viewport which is a feature for blender that i really enjoy render man does have a viewport render but because it's a little bit on the slower side and it is quite demanding on the cpu or at least my cpu it is not usable for the most part with the blender settings you can see that i tried to keep everything fairly similar i will point out that caustics are on they were on for redshift as well but they were off for renderman same amount of gi bounces however as redshift there however for the most part i tried to keep the workflow and the translation from renderman to redshift and renderman to blender the same i didn't go into any material optimization and i didn't change the lights however i did it change one light in particular or two actually the hdri light from both the renderman and redshift version portal lights work a little bit differently in blender than they do in redshift and renderman as with renderman and redshift you can create an environment light and a portal and basically what happens is the light from the environment only will be shot through the portal whereas with blender if you create a portal light it will encourage the rays to go through the portal and try and spend more time rendering the rays that go through there sort of as like a priority however you would have to fully enclose the room for this to work as a light rays will still be present from the surrounding environment light what i could have done to make the room enclosed was put two extra walls on the outside and invert the normal and use back face culling however with this i would have ran into possibly internal bouncing on the room so i would have got secondary rays bounced from those internal walls that you aren't actually seeing from the outside where the camera is and i was starting to think maybe this is too much of a difference between these two rendering engines to make that change so i decided instead to just use the key light that i had outside which is just a small light that is emulating the sun and then i put a larger light on the window and this is acting as the color that we were receiving from the environment light and the other renderers there is a slight difference in the way that this looks but i got it pretty close i think so let's have a look at some renders here's the first still with cycles at 16 samples the render time came in at 32.1 seconds and you'll notice a couple of things the back wall with the wallpaper doesn't actually look too bad under the optics denoising in this one and also the area under the desk there that i was having issues with in redshift has pretty much been alleviated that back corner there where their shadows were is still a little bit chunky but actually not too bad overall the window cell is experiencing a little bit of artifacting however at 32 samples and only two seconds slow you'll see a significant improvement in noise and this is probably the best place to point out why we're seeing these render times the majority of our render time is actually coming from scene translation so about 30 seconds of this render time and each render time is actually getting the scene translated so without that or if that was you know halved or somewhere thereabouts you would see render times at this sample rate much lower and comparable to redshift to be frank going up to 64 samples you see the noise and the detail is starting to come together a little bit better and we're starting to see a little bit more accuracy in our specular highlights on things like the coily cable attached to the headphones and that back wall is starting to look a little bit nicer as well the render time has only gone up by a few seconds as well so you can see that that scene translation is still taking out the majority of our render time there and finally let's have a look at 128 samples you'll see that this has pretty much captured all the detail that we're looking for and things like that headphone cable that i pointed out earlier you'll notice a little bit more saturation in the walls there because it's able to sample those colors a little bit better so as a still this looks perfectly fine i have also rendered up 256 samples and 512 so i could compare those as well because i did and so for redshift and expectedly they come up with a much finer result and with a slightly higher render time interestingly however you'll see once we get to 256 samples cycle starts to make up the time that it was losing on the scene translation whereas redshift's render times is increasing at the same rate that you would expect so while that's positive for stills why don't we have a look at how it looks for the animation and this is where we run into some issues you'll see that optics is not really doing a great job at holding it together once we're animating the biggest problems that we're seeing here are occurring with the floor there and that area in the dark under the desk isn't great either however i would say the most distracting thing unfortunately is the fireflies which are going to be in our focal point occurring on the black keys of the keyboard and on the glasses there to the left and while compositing does cover up some of these issues that floor is just still a little bit too noisy and distracting and the fireflies are obviously going to be very present as they are directly in the middle of what we're looking at so i rendered it up to 512 samples still using optics 2d noise and you see the floor is a little bit better and the other areas that we're experiencing noise in the background have cleaned up a bit but our main issue still is these fireflies in the center of our shot we really don't want those because that is where we want people to be looking and you'll see again that compositing doesn't cover up these fireflies however the background is probably getting i would say within the realms of being reasonable so at this point our only option is to change our denoiser so what i decided to do was use the compositing denoiser for blender which was recommended in the previous video even though i was only really comparing optics for redshift and cycles so here are the versions of this animation with the compositor used as the denoiser and i would say 128 samples any still frame from this denoiser looks excellent and would be fine for final production however in the animation while it's cleaned up those fireflies on the keyboard the floor is still flickering quite a bit so let's increase the sample rate to 512 and instantly you'll see that the majority of the noise in the background is pretty much alleviated and we're still getting some fireflies on the glasses there but the keys are pretty clean and it gets a little bit better obviously once we add the compositing in so i wanted to do one more sample rate just to make my point very clear here so i added 2048 samples in using the compositing denoiser and you'll see that it hasn't fixed our issues really that much actually in some ways it's kind of worse the keys there actually have a few more fireflies and some frames and the floor has gotten quite a bit worse the glasses actually seem to be a little bit worse as well surprisingly now this isn't necessarily what i would have expected but i did want to show it the reason i want to show it is because with this comparison my objective is to show the differences between the renders and not just tell you which one is the best this is what happened in my experience with the scene when translating it from maya to blender now can i get the scene to be cleaner at a lower sample rate in an animation absolutely and i do plan on doing that in a separate video but for now what i wanted to do was just show the direct comparison when translating all the exact same things as best as i could between each renderer one thing i'll also just quickly point out here when i did the compositing denoise i actually changed the colors on a couple of the materials just slightly the main thing being the chair and the table the way that blender is multiplying colors versus noise is a little bit different the calculation must be different between maya and blender i'm not sure why this is i'm using linear color space in both software but for some reason it made the seat a lot more desaturated in blender i'm not quite sure why that is feel free to let me know in the comments if you think you have a good idea so here are all the renders side by side and if you wanted me to talk about winners and losers you've come to the wrong place unfortunately i don't really think that any one of these renderers is better than the other i think that they could all perform at the exact same level given the opportunity because each scene is going to be unique you're not going to be able to just brute force and fix everything just by increasing the sample rate you will require slightly different things to be done in each scene that you work on if you're unwilling to adapt you're not going to be able to get jobs in the industry because if you are say married to redshift and you go to work at a say disney studio do you think they're going to be using redshift or do you think they're going to be using renderman the same goes for cycles if you are only planning on ever using blender to do anything that's fine but you are going to find it slightly difficult in future to get hired versus someone that is using the industry standard software that may already be integrated in certain studios now i'm not saying that this is the case for all studios plenty of studios are starting to use blender but your versatility is going to be a strength all right that is it for this comparison and that is the word that i want to just touch on here this is just a comparison this isn't incredibly scientific it is just me doing it it is me translating it i'm not going to go into all the very very technical things that would be required to make this a scientific comparison between all the renderers in the previous video a select amount of people had a lot to say about my bias i was pretty clear in that video about biases that i may or may not have and i think i've been clear enough in this one so i'm not going to discuss it any further in the comments if you do have some criticism about the video please feel free to share it but i'm not going to tolerate being attacked just bear that in mind i don't think many comparisons out there actually go into specific scenes most of them are very simple setups so i wanted to do something a lot more in-depth and it's a scene that i'm very familiar with because i spent a lot of time creating it and working on it if you like the scene and you'd like to do something with it yourself it's actually going to be available for purchase on our gumroad patrons will get a 40 off unless you've been a patron for six months or more in that case you get it for free so patrons coming up to six months maybe just hold off you'll get it after the six month point other than that um if you're an existing patron you don't want to wait that long you have the 40 discount i'd also like to point out that there's going to be upcoming updates to blender that will solve possibly some of the issues that i talked about there as well as renderman24 coming out and also coming out for blender so look out for that in the very near future maybe i'll revisit this comparison for the renderman 24 performance boost that hopefully we will see xpu as far as i know is not going to be out upon release however it will be available sometime in 2021 i believe hopefully you found the video enjoyable or informative or interesting or all three so make sure you click subscribe and the notification bell so you know whenever we create new content because subscribing just is not enough nowadays so thanks for watching and i will see you in the next video you

Channel: Small Robot Studio

Views: 46,988

Keywords:

Id: 6v7ZuFXPPQg

Length: 24min 10sec (1450 seconds)

Published: Wed Jan 20 2021