Guest tar Posted March 4, 2015 Share Posted March 4, 2015 . Especially now there is tougher competition from Solidangle and Pixar's RIS is really good, Mantra has some catching up to do. It would be best to send these tests into SideFx - it is the very first time I have seen written that Arnold and PixarRIS is ahead of Mantra. Better yet - post your results and tests to Odforce too Quote Link to comment Share on other sites More sharing options...
sebkaine Posted March 4, 2015 Author Share Posted March 4, 2015 (edited) Thanks for your feedback tinyparticle ! I agree that Maxwell is the fastest engine if you judge it by what it really compute compare to the others. Arnold is not an option because i find it far too expensive for what it offer in reality compare to Mantra, but they have the absolut dream team of Rendering engineers there so i am pretty sure it works smoothly ... RIS / Katana is cool but like arnold far too expensive ... and using RFM after using MTOR for 3 years is not an option for me. Now the next client for testing is clarisse, i'm waiting for a 2.0 trial to start testing ... I will post the clarisse vs mantra test ! I think that Houdini with his stellar integration of openVDB could investigate in the direction of voxelisation like Maxwell do. i don't know the R&D part , but i have the feeling that VDB could be use to store lighting / rendering data in a very efficient way. Point cloud are cool but with large scene you ended up with 20GB .ptc to read / frame. I am quite sure that a VDB volume could store vector / floating data of a 3d scene in the most efficient / fast / light way you could imagine ... as a side note what are the VEX command that allow you Read / Write operation inside a VDB object ? Edited March 4, 2015 by sebkaine Quote Link to comment Share on other sites More sharing options...
Erik_JE Posted March 4, 2015 Share Posted March 4, 2015 http://forums.odforce.net/topic/16507-corona-render/page-2#entry101045 This post by old school is pretty great on how to get the most out of mantra 1 Quote Link to comment Share on other sites More sharing options...
sebkaine Posted March 4, 2015 Author Share Posted March 4, 2015 Fantastic Post Erik, Thanks for the links ! Quote Link to comment Share on other sites More sharing options...
Skybar Posted March 4, 2015 Share Posted March 4, 2015 It would be interesting to see Octane tackle this, I imagine it would work out pretty well. Have you tried out Arnold in Houdini as well for comparison recently? You can download it on their site for free, although it renders with an ugly watermark it would be interesting to compare the noise/time. Quote Link to comment Share on other sites More sharing options...
Mzigaib Posted March 4, 2015 Share Posted March 4, 2015 I have make a last try by decreasing pixel sample and augmenting light sample quality , and i gain a little sharpness. Here is a new render in 1080P => 1h14 of computation - pixel sample 4*4 - min sample 20 - max sample 60 - noise level 0.004 - PBR + Env Direct Lighting (sample quality 1.5) + Portal Geo + PhotonMap (filter sample 1 / Distance Threshold 1) I think i a have reach the max i could do for reducing render time without affecting quality too much. Now to go around 45 min per frame i would see 2 way ... - replace direct lighting by ambient occlusion - find a better way to cache GI ( ptc ? voxel tree ? other ? ) At the end Mantra is very solid, and the flexibility it offer is still very cool. In this regards better than Maxwell I would still suggest trying to decrease pixel samples even more like 3X3, I don't see in this particular case problems with anti aliasing like very small details or displacement or even detailed reflections so you can keep the pixel samples low, and bump up the max ray samples to say at least 100 and up, also you mentioned that you are using besides a portal light a environment light I would ask if it is really necessary since it is a indoor scene, with an environment light plus your portal light you are going to increase the chances of more noise, I wonder if removing it can smooth things a little bit but I can be wrong. I am really surprised about how you are saying that Maxwell has an edge on Mantra on the speed matter because some co-workers of mine used Maxwell for some lookdev work but never to finals it was not viable because it's longer render times, remember things change when you go from a still render to an animated camera. I am also curious about the results with Maxwell can you comment more about it? Maybe some render times? I did this exact same test with exact same scene about 3 years ago Houdini vs Maxwell vs Arnold and I have gone pretty extensive pretty crazy with the tests. And almost every test, Maxwell came on top. I did not cut any corners and did not use any bias like photon mapping. Mantra was a little ahead of Arnold by then. I did a similar test using a different scene recently and I saw that Maxwell again came on top but this time Arnold was pretty close to it and Mantra was lagging behind. I have also experienced the same errors with ray variance. I am not sure about the source of the issue but I believe there is a lot of room for improvement for Mantra. Especially now there is tougher competition from Solidangle and Pixar's RIS is really good, Mantra has some catching up to do. Yeah, like I said I am curious about how much behind Mantra would be because that was not what I've witnessed, can you comment about it? It would be best to send these tests into SideFx - it is the very first time I have seen written that Arnold and PixarRIS is ahead of Mantra. Better yet - post your results and tests to Odforce too Yes definitely! You could also mention to them the ray variance problem. Quote Link to comment Share on other sites More sharing options...
sebkaine Posted March 4, 2015 Author Share Posted March 4, 2015 (edited) I would still suggest trying to decrease pixel samples even more like 3X3, I don't see in this particular case problems with anti aliasing like very small details or displacement or even detailed reflections so you can keep the pixel samples low, and bump up the max ray samples to say at least 100 and up, also you mentioned that you are using besides a portal light a environment light I would ask if it is really necessary since it is a indoor scene, with an environment light plus your portal light you are going to increase the chances of more noise, I wonder if removing it can smooth things a little bit but I can be wrong. I am really surprised about how you are saying that Maxwell has an edge on Mantra on the speed matter because some co-workers of mine used Maxwell for some lookdev work but never to finals it was not viable because it's longer render times, remember things change when you go from a still render to an animated camera. I am also curious about the results with Maxwell can you comment more about it? Maybe some render times? Yeah, like I said I am curious about how much behind Mantra would be because that was not what I've witnessed, can you comment about it? Yes definitely! You could also mention to them the ray variance problem. Thanks for all the tips Michel ! Well for Maxwell, don't forget that i mention it is the fastest "for what it computes" In fact if you are comparing with other engine 2 bounce of diffuse is often the norm before thing start to be very slow. Maxwell doesn't limit the bounce number. So i would say that for 1080P Maxwell will be around 2h30-4h00 per frame, which is definitly slow for us simple mortal who compute stuff on our single PC. But at work with a nice farm 2h30-4h00 is quite decent. The area where maxwell suck is sss for the moment. It's very slow especially if you compare it with .ptc approach. i have send the scene to a friend which is black belt Maxwell Ninja, i ask him to get the best result in 1080P for less that 1h30 per frame. i will post the result soon if he succeed. I have ask to a friend to send the stuff in v-Ray. he get this for 14minutes in 1080P of render with Irradiance Map / Lightcache combination. V-Ray is still a killing beast for versatility ! I haven't try Arnold cause i will never be able to buy license for my own buisness ... Cheers E Edited March 4, 2015 by sebkaine Quote Link to comment Share on other sites More sharing options...
zoki Posted March 4, 2015 Share Posted March 4, 2015 great thread one thing worth mentioning though in production(VFX, FIlm, Animation) this kind of sampling would be rarely used as we mostly render moving images with motion blur we can get away with a bit more noise. This kind of sampling is very useful mostly for arch viz and industrial desing type of rendering. I believe there is some old thread somewhere where jason posted some guidelines from R&H, or was it DD? But still nice research here thanks Quote Link to comment Share on other sites More sharing options...
sebkaine Posted March 4, 2015 Author Share Posted March 4, 2015 good point zoran , this is especially true when you use Maxwell and that the next SL for some less noise is gonna cost you 2 hours ... I am still curious on how it is possible to get a similar workflow like vray in Mantra. Generating the irradiance and baking it , then read it ! Photon Map looks to be the official way to bake in Mantra, but i 've seen some .ptc experiment in the past. But if you have any suggestion to reduce rendertime and get closer from 30-40min per frame in 1080P it would be cool. Quote Link to comment Share on other sites More sharing options...
Mzigaib Posted March 5, 2015 Share Posted March 5, 2015 By default photons bake indirect illumination but you can also bake irradiance and direct illumination thou I've never tried, maybe it is worth give it a shot. Quote Link to comment Share on other sites More sharing options...
Guest tar Posted March 5, 2015 Share Posted March 5, 2015 OctaneRender v2.16 standalone demo app . Path Tracing Kernel, Glossy material, Environment light Gtx 980. Limited Res 1002 x 602- OsX * Warning * operator experience with Octane < 10min Total time ~17.5 min Quote Link to comment Share on other sites More sharing options...
tinyparticle Posted March 5, 2015 Share Posted March 5, 2015 (edited) @sebkaine: I think you can register at Solidangle web site and test drive arnold for free but with watermarks. It will serve its purpose for comparison. Edited March 5, 2015 by tinyparticle Quote Link to comment Share on other sites More sharing options...
sebkaine Posted March 5, 2015 Author Share Posted March 5, 2015 @Michel Thanks for your feedback @Marty The light looks very good in Octane and the speed is very good. I have also include it because i love the shading and global logic which is very close from Maxwell. But i have really a big problem to go with full GPU / CUDA engine. Redshift is also a serious client it's V-ray on the GPU and the speed look blazing fast. But again be unable to lunch renders just because NVidia has publish a new driver that break the engine compatibility is a guess not pure science fiction. @Tiny Particle i should do that, Quote Link to comment Share on other sites More sharing options...
Guest tar Posted March 5, 2015 Share Posted March 5, 2015 ..and here is Arnold - it looks quite fun to play with, IPR view was a bit crashy but was solid rendering out to MPlay, really needs more skill than my 25min of experience with it . All the noise is in the indirect diffuse/specular, quickly read the excellent user guide and tried upping the samples on the Skydive light to 40 but was still quite noisy. Pixels 1004*600- to match Octane's demo frame size limitations. 21.5 min OsX, 2 x 3.33 6 core Xeon Arnold Scene in H14.0.201.13 https://www.dropbox.com/s/rsscp92sf41jblz/natural_history_EDIT5_Arnold.hip.zip?dl=0 Quote Link to comment Share on other sites More sharing options...
sebkaine Posted March 5, 2015 Author Share Posted March 5, 2015 (edited) @Marty => Quite a good synchro ! Ok here we go with Arnold: 720P => 34min17s so around 1h15 for 1080P. Settings : - camera AA : 8 - Limit Diffuse : 2 - Limit Glossy : 2 AiSky : - Emit Diffuse only - Resolution :1000 - Samples : 4 - Max Bounce : 999 So i would say that it look to be a little faster than mantra, but the noise in this image is still high compare to what i have set in mantra. But you lost all the flexibility of Mantra. (Deep Shadow for volumetrics etc ....) As you limit the number of bounce it also perform faster than Maxwell but the fact that Maxwell save .mxi that allow you to don't restart from scratch if you need more sample is a killing features. for 1 arnold user license + 10 nodes you are at 13000euros with maintenance. For 13.000 euros you have 15 user license and 150 nodes of Maxwell ... well if i would have to choose a Brute Force engine my choice will still be Maxwell even if render are longer ... Now i'm gonna throw 34m17s test into maxwell just for comparing ... Edited March 5, 2015 by sebkaine Quote Link to comment Share on other sites More sharing options...
zoki Posted March 5, 2015 Share Posted March 5, 2015 hi this is just basic maxwell setup phsycial sky default basic material all throughhoudini 17m 43 sec Quote Link to comment Share on other sites More sharing options...
dennis.albus Posted March 5, 2015 Share Posted March 5, 2015 (edited) Hey, here are a couple of things that I found helpful when working with mantra. 1. Set your min reflection ratio to 1. This helps in reducing artifacts because Houdini is not trying to be smart by reducing your secondary samples by an arbitrary amount (in this case 0.1) 2. Min Ray Samples needs to be higher than 1 if you want to get rid of artifacts but should not go much higher than 4 usually. It needs to be higher than 1 so Mantra has a certain amount of information to decide whether to sample more or not. 1 sample is not enough for that. 3. I found that upping the Pixel Samples is often more successful than increasing secondary ray samples. I suspect that this is due to the way mantras random number generator works. With a new pixel sample a new SID (sample ID) is generated which helps decreasing correlation artifacts and results in a more evenly spread sampling noise. In most cases you need to increase your pixel samples anyway when starting to work with textures and especially when Motion Blur or DOF enter the mix. 4. Primary noise that stems from insufficient 'shadow samples' can only be efficiently reduced by increasing light samples. Otherwise you have to go crazy with the rest of the sampling parameters. Try to use the diagnostic planes to identify the type of noise you are getting. The recent splitting of direct and indirect ray samples can also help to get rid of 'primary' noise. I hope this is helpful. One more thing: I believe the Maxwell voxelization step is just the generation of the raytracing acceleration structure (like mantras kd-tree) and has nothing to do with interpolation techniques a la Brickmaps. cheers, Dennis Edited March 5, 2015 by dennis.albus 4 1 Quote Link to comment Share on other sites More sharing options...
sebkaine Posted March 5, 2015 Author Share Posted March 5, 2015 (edited) @Dennis Many thanks for all those great advise, i will investigate this further. I think the idea of : - pushing the light sample only (already mention previously) - uping sample to 8*8 (arnold recommanded value for final render) - tweaking min / max / noise accordingly is a good nex step. Here we go with Maxwell Quality ... 720P = > 35min of rendering SL reach 14.75 Benchmark 325 i think we need to go at leat to SL 17 for good outpout. It's interesting to see that Maxwell outpout and Mantra outpout are very close. Mantra in all those test is one of the best compromise beetween quality / flexibility / cost If it was only me i think it would be my first choice for the moment, the limitation and default are extensively the result of ignorance / bad understanding of it. but the more you dive into it the more you get ... Shame that lighters don't want to work / are afraid to work in H cause it would be my first choice at the moment ... The bad point are that all this power is pretty rough and need some understanding and to be display in a user friendly way. I am also quite sure that a tool like houdini with kickass Vdb / Kickass ptc should come with better baking / reading than only photon map. Edited March 5, 2015 by sebkaine Quote Link to comment Share on other sites More sharing options...
zoki Posted March 5, 2015 Share Posted March 5, 2015 excactly "Mantra in all those test is one of the best compromise beetween quality / flexibility / cost" Quote Link to comment Share on other sites More sharing options...
sebkaine Posted March 5, 2015 Author Share Posted March 5, 2015 i just ad a comparision : Quote Link to comment Share on other sites More sharing options...
Recommended Posts
Join the conversation
You can post now and register later. If you have an account, sign in now to post with your account.
Note: Your post will require moderator approval before it will be visible.