Memory use?
- Thread starter Innerdreamrecords
- Start date
That is totally insane! I feel your pain too. My most recent fridge render took some 90000 secs!!!That is ONE day, for a stinking 480 X 320 image. What's the deal???
Andreu,
I'm not saying that you don't know what you're doing, but before posting something like " That is totally insane! " and jumping in the wrong band, maybe you could think twice ?
90000 secs ? Something is wrong ? Possibly ! How do you think we can help or Martin can resolve a possible issue without having any details on your scene and on what you are doing with it ??????
Post your scene and/or give full details on it and then we can see what's wrong, whether it's you with crazy settings or the software ...
Sorry for the rant but you're on the wrong league and pushing the limits ...
Hey Francois,
I agree I should attach a description of the setup in order to determine what's causing the high render time. As for the rest, I don't think my post is that caustic for you to react in such an inflammatory way. The "that's insane" phrase refers to the original post by IDR.
So here's my setup. I hope it's not too crazy for ya.
Camera: No tags. No light. Edge antialiasing. Output: 500 X 550.
Setup 1: 10 area lights @ 2 samples each, raytrace + transparency for each, geometry on, 100 caustic photons each. Render time: 86000 secs.
Setup 2: 6 area lights, all raytrace + transparency. One of them at 12 samples, no geometry, 10000 photons. The other five, 1 sample each, geometry on, no photons. Render time: 76000 secs.
Setup 3: 1 single point light with 1 sample. Render time: 10000 secs.
System: iBook G4, OS X 10.5.4, 1.42 GHz PPC G4, 1.5 Gb RAM. Cheetah 4.6.2.
I agree I should attach a description of the setup in order to determine what's causing the high render time. As for the rest, I don't think my post is that caustic for you to react in such an inflammatory way. The "that's insane" phrase refers to the original post by IDR.
So here's my setup. I hope it's not too crazy for ya
.Camera: No tags. No light. Edge antialiasing. Output: 500 X 550.
Setup 1: 10 area lights @ 2 samples each, raytrace + transparency for each, geometry on, 100 caustic photons each. Render time: 86000 secs.
Setup 2: 6 area lights, all raytrace + transparency. One of them at 12 samples, no geometry, 10000 photons. The other five, 1 sample each, geometry on, no photons. Render time: 76000 secs.
Setup 3: 1 single point light with 1 sample. Render time: 10000 secs.
System: iBook G4, OS X 10.5.4, 1.42 GHz PPC G4, 1.5 Gb RAM. Cheetah 4.6.2.
Last edited:
Andreu,
Don't sweat Francois he's a fanboy, he seems to think you shouldn't voice you're real opinion, maybe he lives in a communist country? :wink:
There is something not quite kosher going on I suspect.
Don't sweat Francois he's a fanboy, he seems to think you shouldn't voice you're real opinion, maybe he lives in a communist country? :wink:
There is something not quite kosher going on I suspect.
It's possible there's no reason to use more RAM. 500MB is quite a bit.
If you go by the basic rules of rendering time -- you need to watch out for things that blow out rendering time geometrically.
To raytrace a scene you basically perform a recursive operation for every pixel in the screen (more if you anti-alias or render subpixels for some other reason). Ray depth increases the amount of recursion.
Every light source adds another pass. Area lights are basically implemented as multiple light sources (with some clever shortcuts)
Caustics multiply the number of passes. (You're talking about photons; photons imply caustics.)
Radiosity multiplies the number of passes.
So if you take a simple scene that renders in, say, 30s, and then increase raytracing, add a bunch of area lights, add caustics, add radiosity, increase ray depth -- well you are multiplying your 30s by (say) 20, 10, 5, and 3 for a scene with 10 area lights and a fair bit of glass/reflective surfaces.
30s x 20 x 10 x 5 x 3 = 90,000s.
If you go by the basic rules of rendering time -- you need to watch out for things that blow out rendering time geometrically.
To raytrace a scene you basically perform a recursive operation for every pixel in the screen (more if you anti-alias or render subpixels for some other reason). Ray depth increases the amount of recursion.
Every light source adds another pass. Area lights are basically implemented as multiple light sources (with some clever shortcuts)
Caustics multiply the number of passes. (You're talking about photons; photons imply caustics.)
Radiosity multiplies the number of passes.
So if you take a simple scene that renders in, say, 30s, and then increase raytracing, add a bunch of area lights, add caustics, add radiosity, increase ray depth -- well you are multiplying your 30s by (say) 20, 10, 5, and 3 for a scene with 10 area lights and a fair bit of glass/reflective surfaces.
30s x 20 x 10 x 5 x 3 = 90,000s.
Pod,
I'm sure you know way more than I ever will but that contradicts what Martin told me when I only had 512 MB's of ram. He said that is barely enough to run the program let alone do a complete scene. Also, what would the point be of having 4 GB's of RAM if Cheetah doesn't access it? I have been using this program long enough to know that a scene in my case 1366x768 with 4 area lights and 4 spot lights should not take over 15 hours.
My radiosity was set to 600 samples and I used the default in the radiosity for all the other settings.
I'm sure you know way more than I ever will but that contradicts what Martin told me when I only had 512 MB's of ram. He said that is barely enough to run the program let alone do a complete scene. Also, what would the point be of having 4 GB's of RAM if Cheetah doesn't access it? I have been using this program long enough to know that a scene in my case 1366x768 with 4 area lights and 4 spot lights should not take over 15 hours.
My radiosity was set to 600 samples and I used the default in the radiosity for all the other settings.
Attachments
Last edited:
Well, I hear what you're saying, pod, and I agree, although some part of me gets flashbacks of the times when I used to leave the computer working on a Gaussian Blur all night long. I swear it has happened
As for improving render times, I don't think it is as crucial that they be rock-bottom low when you're outputting the final scene. But I do think that long render times interfere in the creative process. So, I was just going through all the possible causes for slowdown, and I noticed the antialiasing and oversampling options in the camera object. I never touch those settings from the default, but I just lowered min. samples to 1 x 1 and max. samples to the same, and that changed things by a lot, although the quality is just good for testing and checking out things.
I would benefit from having or making myself a sort of reference chart of things to go through when render times go bad. A sort of back to the basics check out list, where you build up safely from settings that are proven to perform and output at a specific level.
BTW, that is an amazing render, IDR! Let me tell you, 54K sec is nothing!
As for improving render times, I don't think it is as crucial that they be rock-bottom low when you're outputting the final scene. But I do think that long render times interfere in the creative process. So, I was just going through all the possible causes for slowdown, and I noticed the antialiasing and oversampling options in the camera object. I never touch those settings from the default, but I just lowered min. samples to 1 x 1 and max. samples to the same, and that changed things by a lot, although the quality is just good for testing and checking out things.
I would benefit from having or making myself a sort of reference chart of things to go through when render times go bad. A sort of back to the basics check out list, where you build up safely from settings that are proven to perform and output at a specific level.
BTW, that is an amazing render, IDR! Let me tell you, 54K sec is nothing!
Last edited:
Andreu,
The original post was "... seems insane ..." and with no details.
Your post was "That is totally insane" and without quotes.
If you were refering the original post, how can you replace "seems" by "totally" without having the details ? And you were not giving details either on your scene on an even longer rendering time !
Sorry if you feel I reacted in an inflammatory way. That was not the intention nor the point !
If you want to complain about something, that's fine ! But you can't do it without fully detailing why, so others can help.
Regarding your settings, that's what I was expecting : a too high number of area lights. You now have the right explanations from Pod and should be able to reduce your rendering time.
Apology accepted.
Sometimes it's fun to just complain for the sake of it, ain't it?
Well, as a Mac user I have learned a very valuable lesson from The Macintosh Bible. To paraphrase, "if you can't do something, don't blame yourself, blame the software." This statement may seem like an escapist way to approach problems, but it often reflects the state of affairs about issues and shortcomings when dealing with program functions.
More to the point, in Cheetah3D, if I can't place ten area lights in a scene, with no extra and fancy settings on the camera, and expect the render time to be relatively moderate, then it is the render engine, not my settings, who is to blame. Ten area lights is what you get any day of the week in any architectural scene with a few windows. Also, remember, I had not a single tag in the camera or otherwise. So, here I am with IDR in thinking there is something fishy going on. I'd like to find out too, but I am not going to blame my "crazy" settings, because in my experience, more often than not this kind of issue has to do with something I'm neither responsible for nor capable of improving.
Francois,
If you look at my post and read it correctly you will see that I posted a jpeg of the memory use. The jpeg shows clearly that the program is not using all of the available ram. That is the question and the issue.
I know you think you are helping to preserve the integrity of Cheetah but in fact as far as I know you don't work for the company and you come off as the unofficial policeman. Why not simply say "could you give some more details so we might be able to help?" Or better yet if you can't help stay out of the conversation.
If you look at my post and read it correctly you will see that I posted a jpeg of the memory use. The jpeg shows clearly that the program is not using all of the available ram. That is the question and the issue.
I know you think you are helping to preserve the integrity of Cheetah but in fact as far as I know you don't work for the company and you come off as the unofficial policeman. Why not simply say "could you give some more details so we might be able to help?" Or better yet if you can't help stay out of the conversation.
There were no excuses in my answer. And I will not even comment the other lines of your post !Apology accepted.
Martin
0
Mmm, why should it run faster if it uses all RAM? Would Safari be faster if it would use all available RAM? OS X apps usually take the RAM they need and keeping two or three copies of the same scene in RAM to fill it up wouldn't make anything faster. But it would be nice if it would be that easy.
The main source of high rendering times are currently a) blurry reflections in combination with area lights and b) the irradiance cache under high resolution renderings or high geometric details.
a) will be addressed with the new material system in 5.0. b) will be probably addressed in 5.1 or 5.2
In the kitchen scene you send me some time ago you had many materials which used blurry reflections. But you used only a angle of 1° for the reflections angle. You can do that of course but the effect will be hardly noticeable in the rendering and the rendering time increases seriously. Especially if you have hugh area lights in the scene. So avoiding blurry reflections with a 1° should already help.
Bye,
Martin
Martin
0
The speed difference between these two settings is very very strange. It should be much bigger. Rendering your refrigerator scene with a single point light shouldn't take that long. How many polygons do you have in your scene. That almost looks like OS X started swapping data to the HD.
I'm very sure Cheetah3D can render Setup 3 in a small fraction of the 10000 seconds on my Intel iMac.
Bye,
Martin
Then what is the purpose of more ram? When we talked about this before as I mentioned you said that Cheetah barely runs on 512 MB's and according to the jpeg that's about all it was using on this current scene. So, is it more processor speed that is needed? Right now I'm confused.
As far as the settings for the blurry reflections that is the look I was after. I don't want to have to tailor the look of the rendering to increase speed if I don't have to.
I'm talking about 500MB of (physical) RAM for C3D itself, not the computer as a whole.
512MBs of RAM (for the computer) is barely sufficient for C3D because Mac OS X needs around 300MB to run without paging to hard disk. So if C3D needs 500MB (say) then it will be in deep trouble using 200MB of free RAM.
C3D can only use 2GB AFAIK (it's a 32-bit program).
Just some random results from very quick test renders:
1 20-sample area light, ball, plane -- 3s
as above, add one area light -- 5s
as above, glass material (transparent, reflective) on ball -- 22s
as above, + transparency for shadow -- 25s
as above, + ambient occlusion -- 25s
as above, with radiosity instead of ambient occlusion -- 40s
as above, with procedural wood texture on the plane -- 40s
as above, with color anti-aliasing mode -- 63s
as above, with caustics -- 64s
as above, camera closer to glass ball -- 200s
as above, subdiv on glass ball -- 260s
It's not hard to blow out render times if you crank up expensive options or quality settings. All I'm doing here is rendering a ball on a plane with two lights at 640x480.
Take a scene which takes 60s to render with basic lighting at 640x480 -- the kinds of scene you're talking about are beyond that I'd say -- then render it at 1280x1024 (say) -- that's 4x the rendering time right there -- now add a decent lighting setup, rendering options, and complex materials (as you've seen that increased this scene's rendering time by ~70x) and you're at 4h.
Now add the fact I'm doing these tests on a 2.16GHz MacBook Pro.
Last edited:
I was intrigued by what you said here, so I went back through the scene. There was an extra area light with six samples inside the fridge. Also, some of the shiny materials had a reflection angle of 1 or 2. I wonder if that would make a difference. I can also send you the file if you think that would be helpful.
But in any case, last night I was testing some bare bones scenes, and this is what I got. I create a cube. No lights. Camera light on. All default settings. Render time 8 secs. I don't know if that is a high render time or not, but it does not seem particularly fast at first sight. It might be my slow iBook. I wonder what you get on an Intel iMac.
Last edited:
I'm glad you took the time to do some testing of your own. The results are illuminating.
As for the quote, I have noticed that the render does not get bogged down across the board, but rather it seems to hit a series of bottlenecks that slow down the overall render time. In particular, transparent objects seem to slow down things. Reflections seem to be fine in terms of render speed. Any comment on this?
Just to follow up so some might get some knowledge. Here is the same scene without the four area lights replaced with two point lights. I rendered it slightly smaller but it was under two hours compared to about 16 hours with the area lights. I personally don't think it looks as good but hey, you get what you pay for.
