Chaos Corona Forum
Chaos Corona for Cinema 4D => [C4D] General Discussion => Topic started by: tennet on 2017-11-06, 15:35:23
-
I'm finding Team Render to run very slow compared to rendering on just one workstation (huge difference). I was using one Mac Pro (8-cores) and one Windows 10 PC (28-cores). I "Team Rendered" a very simple scene (a background wall with just a light on it) and it starts to "build up" the image from the bottom, very slowly (almost like bucket rendering, kind of..). When using the PC alone, then I can have a final render ready in 5-10 minutes (4K resolution).
Any settings I can change or is this already a "known issue"?
-
As I could say, is that team render does not support noise limit option so only passes limit (didnt try time but I suppose it works fine), and niether do denoising. I would like to know also is that a known problem.
-
Any info or help regarding this subject? I just installed the latest RC4 Daily Build for C4D (on both the Mac and PC) and it is still rendering extremely slow when using Team Render. I am starting my render from the PC and on the Mac I have the C4D Team Render Client running. The render Client shows that Corona RC4 is loaded. When starting the Team Render on the PC it also says it uses 2 nodes... but its very very slow. The PC alone renders lighting fast, if compared.
Any tips or ideas if I am doing something wrong? Is the Team Rendering supposed to work with (and speedup!) IR or only meant to be used for "Final Output" (I was testing it for a final render, not IR)? Should the Team Render Client be used on the nodes?
Thanks!
-
Also, I get completely different results when using Team Render compared to rendering on a single machine. The Team Render looks blown out and the textures seems to be mapped all over. Are there any settings I am missing? My Corona settings are set to ""Automatic" for the Team Render options..
-
Also, I get completely different results when using Team Render compared to rendering on a single machine. The Team Render looks blown out and the textures seems to be mapped all over. Are there any settings I am missing? My Corona settings are set to ""Automatic" for the Team Render options..
Not able to duplicate so far - can you share the scene, and examples of your results? Are you using a camera, and does it override any of the settings? Is this "TR to PV" or is it using the TR Server to submit the job?
Thanks!
Tom
-
(And of course the usual question - are you sure you have the same version of C4D, of Corona, of TR, and of all plugins, on every machine?)
-
Thanks! I will collect some screenshots and share the scene with you. This was using "Team Render to Picture Viewer" and then it says in the bottom of the window "rendering using 2 nodes".. I am also running the same version of C4D R18 on both Machines and I have installed the latest Daily Build RC4 on both systems (1 mac, 1 Win10 PC). Will get back to you soon with screenshots and example scene.
-
We noticed the very same issue here.
We tested corona with 3 Machines.
All on Cinema4D R19.053 S and Corona build B1. To my knowledge, these are both the most recent versions.
Scene rendered was Coronas "Release Scene V2".
The picture viewer shows rendering from the bottom (in some sort of "scanline fashion") to the top. Very slowly.
I attached a screen shot for clarification.
Many thanx in advance!
-
9999 passes doesnt seam like a real number to me.. try to put something like 15-30 passes and add the denoise option for later! so you can denoise in post..
i do a lot of teamrendering with a total of 160 cores (5x dual xeon 32cores) and for production quality its a lot faster then rendering on only one 32core dual xeon workstation...
check your lights/mats/multipass & rendersettings before teamrendering and you should be fine in terms of required time.. and ofc for me too teamrendering does "scanline" render from bottom to top but as soon as its up there the rendering got the same quality as when running the same number of passes on one machine
-
Thanx.
The file is Thomas Vournazos free sample scene I found here: https://corona-renderer.com/resources/models (https://corona-renderer.com/resources/models). Not official per se but I think we can presume its settings are mostly accurate. (Not that I tried changing it - to no avail).
The number of passes shouldn't make a difference and was set by Thomas for testing purposes I presume. (I guess he could have filled in 0 for that matter).
Anyhow, it doesn't make a difference it seems. I tested it with 20 passes and had the same issue.
To my knowledge, de-noising doesn't work with TeamRender at the moment. To be safe, I tested it with the de-noiser off and on. With the same result in both cases. I.e. the issue still persists.
The "scan line effect", for me at least, seems unacceptable as it cancels out one of the major benefits of a progressive renderer. That's why I very much hope so that this is a bug.
In addition I don't get even close to the speed boost I should get from tripling my render cores.
-
And I guess I should apologize at some point, for hijacking the thread somewhat.
-
you are right, the "scanline" methos isnt as good as the progressive method but my workflow usually uses 1 machine for testrenderings in low res on normal progressive view and when im done adjusting, the final rendering in high res is done via teamrender for the amount of passes i require. i usually run for denoising the option "store data for later" and when the rendering is done i save as cxr and open in seperate corona image editor for denoising and lightmix adjustments, bloom and stuff.. and what i tested, sure u see the full image first on progressive mode but at the end you get the same result with teamrender just faster as compared to only 1 machine..
-
It isn't rendering in scanline :) What you are seeing is the effect of the limit of the data packet, which limits how much data the node sends back at once. This means the image builds in blocks, as it receives packets of that size, though on the node regular progressive rendering is going on as normal. Corona simply can't do scanline rendering at all under any conditions :) As a note, once rendering is complete, all the data packets from the nodes will be sent over, so you still get all the results of the (progressive) rendering from all the machines, just received in batches based on the packet size.
You can adjust the size of the chunks in the Render Setup dialog, Corona, Team Render settings by changing it to Manual and entering a size for the maximum packet to be sent at once (see screengrab) - larger values will send larger packets, meaning the image will build up in larger chunks. Again though, it has no effect on how things are being rendered on the workstation or node, and only affects how that data is transferred between the two.
The other parameter in there changes how often data should be collected from the TR nodes (longer will mean a longer waits for results from the nodes to come in, but they will have had more passes done since they'd have run for more time).
-
I should add, larger packet sizes may cause more network congestion (as far as I know), especially depending on how often they are sent.
-
Thank you Tom.
Forgive me if I stick with the "scanline" analogy for now.
If I interpret your answer correctly, we will always get that "scanline" effect when rendering with TeamRender. Is that correct? If so, are you guys planing to "fix" that in the future?
@hog0: I realize that there are workarounds to every problem. Me personally, I'd very much prefer not to do this particular one though.
-
You can remove that effect simply by making the packet size large enough to send the whole image at once, I would think (it did in my tests) - but as noted, this could lead to network congestion if you are sending very large packets very frequently from lots of machines.
There isn't anything to "fix" here though as nothing is broken :) The data has to be sent across in chunks of some size... I guess the only way to change it would be to make the packets a whole number of the data size required for the entire image at once, but that would be a feature request rather than a fix. It certainly has no impact on the final image or result or rendering speed.
-
Understood. I'll do some testing on my own.
Thank you Tom.