Quite a few questions \"that need answers\" :)

Message boards : General talk : Quite a few questions \"that need answers\" :)
Message board moderation

To post messages, you must log in.

1 · 2 · Next

AuthorMessage
KEP

Send message
Joined: 12 Mar 06
Posts: 80
Credit: 53,305
RAC: 0
Message 7766 - Posted: 18 Feb 2008, 11:41:52 UTC

Hi

As of tonight, local danish time, I\'ve completed a 112 animations, with a lot of SSS and HD. They all contain trackings. Also they are all 360 frames each, and they are going to be split in 2, 4, 8, 16, 32, 64 and 128 parts. Now this leaves a few questions that I would really like to have answered:

1. How do I create a project? (since they all 112 animations should be viewed seperately but belongs to same project)

2. When submitting the sessions, which RAM meassurement should I rely on? The one stated in the corner when in \"solid mode\" or the one in the corner when in \"bouncing box mode\"? If none of these how do I find out what amount of RAM is needed to render my session without having to render an entire frame (since that takes just days if not weeks)?

3. I suggest that these 112 sessions is going to be handled as sleeping sessions simply because they are going to require around 15 CPU years for the lengthies (according to my best quesses), but how do I make the sessions sleepy sessions, and thereby allows other sessions with high priority to be rendered in reasonable time?

That was about it, and then let me add, according to the memory now, it states that nothing more than ~200 MB RAM is needed per frame, so the first sessions will propably require only 64 MB RAM or as little as 28 MB RAM. So are you up for the challenge everyone?

Regards

KEP
ID: 7766 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
AC
Project donor
Avatar

Send message
Joined: 30 Sep 07
Posts: 121
Credit: 143,874
RAC: 0
Message 7767 - Posted: 18 Feb 2008, 17:30:07 UTC - in response to Message 7766.  

Just a warning: As far as I know, SSS still doesn\'t work correctly for multipart frames on BURP.

It also kind of sounds like you might have some settings turned up higher than you might need -- remember: a longer render time doesn\'t necessarily mean higher quality. A big part of producing good computer graphics is learning to do a lot with a little.

Anyway, that having been said, I\'ve enjoyed your previous animations and am looking forward to seeing the new stuff.
ID: 7767 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
KEP

Send message
Joined: 12 Mar 06
Posts: 80
Credit: 53,305
RAC: 0
Message 7768 - Posted: 18 Feb 2008, 18:33:20 UTC - in response to Message 7767.  
Last modified: 18 Feb 2008, 19:25:03 UTC

Just a warning: As far as I know, SSS still doesn\'t work correctly for multipart frames on BURP.

It also kind of sounds like you might have some settings turned up higher than you might need -- remember: a longer render time doesn\'t necessarily mean higher quality. A big part of producing good computer graphics is learning to do a lot with a little.

Anyway, that having been said, I\'ve enjoyed your previous animations and am looking forward to seeing the new stuff.


I guess you\'re right, but for the exact purpose as I had to achieve here, SSS just made the colors look exactly like I wanted to. However I\'m a bit puzzled as to weater I need to do some subsurfing on my figures or not, but I guess that it will just take only few minutes per animation to do. This time I can also add, there is no background and no textures.

As a preview, I can tell you all, that each animation represents each atom in the Periodic Table. So there you have it, a scientific animation again :)

Edit: Is now uploading \"Hydrogen\" it consist of 2 parts per frame and 360 frames. If it turns out nice once complete, despite the SSS apparently not yet fully supported, the remaining 111 sessions will also be uploaded ;) Also to add subsurface to the Core aswell as the Electrone did not make it appear more smooth so it looked like wasted RAMS. But hey maybe some evaluation should be done after this render and maybe Hydrogen then should be upped again with some changes or maybe everything turns out just fine in the end :) Who knows only time can tell :)
ID: 7768 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
KEP

Send message
Joined: 12 Mar 06
Posts: 80
Credit: 53,305
RAC: 0
Message 7769 - Posted: 18 Feb 2008, 19:41:56 UTC

If I have in the file that I upload to BURP asked for a different amount of split than I\'m able to choose when uploading and submitting my session, then what will happen? If BURP overrides the file prepared and only split maximum 64, then I want bother in the future to waist time fittleling with that menu :)

Thank you.
ID: 7769 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile Janus
Volunteer moderator
Project administrator
Avatar

Send message
Joined: 16 Jun 04
Posts: 4556
Credit: 2,097,282
RAC: 0
Message 7770 - Posted: 18 Feb 2008, 20:05:10 UTC - in response to Message 7768.  
Last modified: 18 Feb 2008, 20:45:43 UTC

Edit: Is now uploading \"Hydrogen\" it consist of 2 parts per frame and 360 frames. If it turns out nice once complete, despite the SSS apparently not yet fully supported, the remaining 111 sessions will also be uploaded ;) Also to add subsurface to the Core aswell as the Electrone did not make it appear more smooth so it looked like wasted RAMS. But hey maybe some evaluation should be done after this render and maybe Hydrogen then should be upped again with some changes or maybe everything turns out just fine in the end :) Who knows only time can tell :)

A few suggestions based on what you write:
1) Rather than using a high-resolution mesh (ie. many vertices in the electron and the nucleus) to make them smooth you should use a low-resolution mesh together with the \"set smooth\" feature on the material (F9->set smooth in object mode) and a low amount of subsurfing (F9->Add modifier). This makes the shading pass much faster, the file much smaller and actually looks more like what you want to achieve.
2) At high image output resolutions (like 1920x1024) OSA has only few advantages unless you have sampled light in the scene. Setting the OSA to 8 is overkill - use 5 or none at all.
3) Instead of sending in 100+ sessions you should consider using the sequencer to create 100+ scenes which are then all rendered in a single session. This will make the render faster and use way less resources. Please note that at the moment both SSS and use of the sequencer limits the rendering to full frames - ie. no split frame rendering. When using the sequencer you can also link and reuse the elements between each scene to save further diskspace.
Alternatively you can think about using just one scene where elements are added and removed over time.
If you do decide to use the sequencer then please set the framesplit to 100x100 in each of the subscenes (to work around a current bug in the client).

If I have in the file that I upload to BURP asked for a different amount of split than I\'m able to choose when uploading and submitting my session, then what will happen? If BURP overrides the file prepared and only split maximum 64, then I want bother in the future to waist time fittleling with that menu :)

The Blender menu settings for splitframe rendering is ignored. It is what you set at the website that counts. That said, it is a good idea to try out different settings ahead of time to figure out what to select on the website.

When submitting the sessions, which RAM meassurement should I rely on? The one stated in the corner when in \"solid mode\" or the one in the corner when in \"bouncing box mode\"? If none of these how do I find out what amount of RAM is needed to render my session without having to render an entire frame (since that takes just days if not weeks)?

A good way of doing this is to start a render and wait until the tracing/shading pass starts (this is when it actually starts showing something on the screen). At this point you can read a memory estimate on the status bar.

A more precise estimate can be found by looking at the total memory consumption. How to find this info differs from operating system to operating system.
ID: 7770 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
KEP

Send message
Joined: 12 Mar 06
Posts: 80
Credit: 53,305
RAC: 0
Message 7771 - Posted: 18 Feb 2008, 20:43:37 UTC
Last modified: 18 Feb 2008, 21:25:42 UTC

Thanks for your advices Janus. I will deffinently look into what you suggested and use it when creating all 112 atoms. Actually that can also give me some sort of knowledge on how to use the sequencer. Even though I\'m not sure what OSA is* (I guess I\'ve used it since you mention it) I\'ll also try and look into that and then upload an entire sequence of all 112 atoms. If I can reduce rendertime by not using SSS, that would actually also be nice, but I\'m not sure if the first atoms nor the last will meet the demand of at least 50 sec. render time per frame. But again I\'ll look into your suggestions and see the difference, once Hydrogen is done rendering, then at least I\'ve something to compare with :)

Also do anyone know how to make an object follow a cirkle more than just 1 round and how to determine the speed? Also is there someway to determine the distance (degress length and so on) between several objects following the same path?

I\'m asking because I\'ve found a lot of crappy tutorials on the internet and none of them are complete. It took me days just to make this tracing work with the hydrogen animation.

*EDIT: It appears that I used OSA as default, so now at least that one is cleared. But if it isn\'t really doing anything, there is no real reason to use it. But hey I guess thats just one more thing to experiment with (the more the merrier) :)
ID: 7771 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
AC
Project donor
Avatar

Send message
Joined: 30 Sep 07
Posts: 121
Credit: 143,874
RAC: 0
Message 7776 - Posted: 18 Feb 2008, 23:04:40 UTC - in response to Message 7771.  
Last modified: 18 Feb 2008, 23:26:03 UTC

Just as an experiment, I reduced the polygon count on the spheres by more than 90% and tried rendering -- it took 25% less time per frame. Then I turned off SSS and tried to adjust the textures so that the colors looked similar, and it took almost 90% less time to render. Here\'s the comparison:

The textures aren\'t a complete match, but here\'s what I did:

  • The emittance was raised to 0.2 to fill out the dark bottom like SSS did.
  • The diffuse reflectivity was increased to 0.9 to brighten the illuminated tops.
  • The specularity was decreased a little to 0.45 to make the highlights slightly less intense.
    (I went a little too far - you can see the difference in the spec)

I\'m sure you\'ll want to adjust these more, but I hope it\'s a helpful starting point... and it should render in a tenth of the time!

EDIT: Oh yeah -- and I think an octree resolution of 64 will render faster than 128 for this animation, if you can find that option.

ID: 7776 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
KEP

Send message
Joined: 12 Mar 06
Posts: 80
Credit: 53,305
RAC: 0
Message 7778 - Posted: 19 Feb 2008, 3:21:02 UTC

Got the Octree, so that I\'ll also look into when trying again. I\'m just glad now that we did not try out with atom 112, since it would have taken some extensive longer time. Also it appears that it will take approximately 48 hours per frame, which is way less than the 110 first guessed.

Also it appears that it will actually look great this Atom. Now with your and Janus suggestions I might be able to recreate them with fewer RAM needed and only 1 file needed for upload :)

All other feedback is aswell welcome, and they will (if not answered) at least be taken notice of.

Thank you!
ID: 7778 · Rating: 1 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile Janus
Volunteer moderator
Project administrator
Avatar

Send message
Joined: 16 Jun 04
Posts: 4556
Credit: 2,097,282
RAC: 0
Message 7783 - Posted: 19 Feb 2008, 7:21:28 UTC - in response to Message 7776.  

Just as an experiment, I reduced the polygon count on the spheres by more than 90% and tried rendering -- it took 25% less time per frame. Then I turned off SSS and tried to adjust the textures so that the colors looked similar, and it took almost 90% less time to render. Here\'s the comparison:.

Amazing. Do you mind if I host your image locally on the BURP server? It is a very useful comparison which may be used as part of a future collection of helpful tips.
ID: 7783 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
AC
Project donor
Avatar

Send message
Joined: 30 Sep 07
Posts: 121
Credit: 143,874
RAC: 0
Message 7785 - Posted: 19 Feb 2008, 7:56:28 UTC - in response to Message 7783.  
Last modified: 19 Feb 2008, 8:15:41 UTC

No problem... the posted version was downsampled a little, the slightly larger original is here if it serves you better. Feel free to use/edit/post.

Actually I\'ve rendered all of the modified 760 (took an hour) if you (or KEP) would like those frames -- got a 30 MB rar archive I can send.
ID: 7785 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile Janus
Volunteer moderator
Project administrator
Avatar

Send message
Joined: 16 Jun 04
Posts: 4556
Credit: 2,097,282
RAC: 0
Message 7786 - Posted: 19 Feb 2008, 8:11:42 UTC - in response to Message 7785.  

No problem... the posted version was downsampled a little, the slightly larger original is here if it serves you better. Feel free to use/edit/post.

Actually I\'ve rendered all of the modified 760 (took an hour) if you want those frames -- got a 30 MB rar archive I can send.

Thanks, I\'ll stick with the comparison for now.
ID: 7786 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
KEP

Send message
Joined: 12 Mar 06
Posts: 80
Credit: 53,305
RAC: 0
Message 7791 - Posted: 19 Feb 2008, 12:12:25 UTC - in response to Message 7785.  

No problem... the posted version was downsampled a little, the slightly larger original is here if it serves you better. Feel free to use/edit/post.

Actually I\'ve rendered all of the modified 760 (took an hour) if you (or KEP) would like those frames -- got a 30 MB rar archive I can send.


It could be kind of nice, if you could send me both the images aswell as the .blend file, so I can get a tip on how you did and maybe use it for future work on my atoms :) Also please tell me how did you reduce the polygons, if that is not obvious once I open the .blend file :)

You can e-mail me at: @gmail.com --> k e n n e t h 0 1 0 9 8 2 (it\'s obvious but infront with this and @gmail.com at the end) also anyone else who has something usefull to say to me can write to that e-mail. It is checked at least once a day if I\'m not on vacation or away for an entire day. Most often it is checked more than once a day :)

Thanks for all your help AC and Janus, now if I may be as free to ask, how has you gotten so good on 3d animation? What is your background? And is there somewhere that I can see some tutorials that is not faulty and really explanitorial, so I can be just as good as you?

Take care!
ID: 7791 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
KEP

Send message
Joined: 12 Mar 06
Posts: 80
Credit: 53,305
RAC: 0
Message 7793 - Posted: 19 Feb 2008, 22:49:14 UTC
Last modified: 19 Feb 2008, 23:04:03 UTC

OK Janus I\'ve now tried to build a Bluewhale DNA string using your advices. It actually seems to save a lot of RAM, since now it is reduced from way more than 2 GB RAM down to 93 MB RAM. Only issue I\'m facing now, is that when making a test render, the computer all the sudden starts building up a huge requirement for RAM. In fact it required at least 700 MB RAM before blender finally stopped working. Now anyone who has any idea, on first of all how there can be such a big difference (maybe I should add that the animation consist of for now 1688 objects) between what is stated in Blender and what it actually requires of RAM to do such an animation? Also can anything be done against the problem or to avoid such a big difference? I am right now thinking that maybe merging/joining all the meshes down to 1 mesh (which it eventually will be in the end, almost) consistent of many meshes? Will it help to split the rendering into more parts than default?*

A good thing however is that the .blend file despite all the many meshes aswell as the camera and lights is only going to be around 17 MB (filesize) which is down from approximately 700 MB (filesize) (at first almost succesfull atempt), so overall a way better file, but I\'m sad to see if I can\'t solve the memory issue that I faces, since I wan\'t be able to submit it for rendering then in the end and also I will not be able to help out myself :(

Thanks for any advices or feedback. Though I\'m asking lot of questions I sure also appreciate your patience with me :)

*Just tested that, it is deffinently not saving any RAM, but it is infact slowing down the sieving. Also it looks like blender when rendering is always using at least 5 times more RAM than stated when you have rendered a testframe :(
ID: 7793 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile Janus
Volunteer moderator
Project administrator
Avatar

Send message
Joined: 16 Jun 04
Posts: 4556
Credit: 2,097,282
RAC: 0
Message 7795 - Posted: 20 Feb 2008, 7:02:50 UTC - in response to Message 7793.  

Can\'t say much without looking at the scenefile.

Mostly memory requirements are based on mesh complexity as well as reflection/refraction and the octree resolution.
ID: 7795 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
KEP

Send message
Joined: 12 Mar 06
Posts: 80
Credit: 53,305
RAC: 0
Message 7799 - Posted: 20 Feb 2008, 11:06:39 UTC - in response to Message 7795.  
Last modified: 20 Feb 2008, 11:38:27 UTC

Can\'t say much without looking at the scenefile.

Mostly memory requirements are based on mesh complexity as well as reflection/refraction and the octree resolution.


Well here is what I used:

844 meshes (8 vertices) with 4 materials. Subsurfacelevel was \"1\" and renderlevel was set to \"6\" to make it round. Before adding subsurface, I selected smooth. To each material there were added a raymirror of 0.3 with a depth of 2!

I then have 842 cylinders with 100 vertices and a depth of 2. These were given only 1 material. Also here I added 0.2 in raymirror with a depth of 2!

I have noticed that there is no octree resolution. Propably because I remember that I accidently pushed the game settings button, so can this be the cause? If the cause is the cylinders, how do I make a round cylinder from a low mesh cube (8 vertices)?

EDIT: If you want to, I can send you the work file. But it appears that it has something to do with the amount of objects. Even using metaballs requires a huge build up in the use of memory :()
ID: 7799 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
AC
Project donor
Avatar

Send message
Joined: 30 Sep 07
Posts: 121
Credit: 143,874
RAC: 0
Message 7800 - Posted: 20 Feb 2008, 18:13:57 UTC - in response to Message 7799.  
Last modified: 20 Feb 2008, 18:21:07 UTC

I was writing this in an email, but it might be useful to others so I\'m putting it here instead. You said you were using a cube with the subsurf modifier\'s Render Levels setting at 6 to make it round. This generates A LOT of faces, which will require a lot of RAM and slow things down a lot.

A subsurfed cube has this many faces:
Level 0 => 6 (no subsurfing)
Level 1 => 24
Level 2 => 96
Level 3 => 384
Level 4 => 1,536
Level 5 => 6,144
Level 6 => 24,576

The number of faces increases exponentially, in this case it\'s 6*(4^L)

The reason for the sudden buildup of RAM is because the Render Levels setting is not applied until you begin to render your scene. This means that when you start rendering, each of your subsurfaced cubes suddenly goes from 6 faces to 24.5 thousand faces.

(Oh, and does ray tracing the reflections really require much more RAM? I wouldn\'t expect that... Oh well, either way RayMirror will definitely take more time.)
ID: 7800 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
KEP

Send message
Joined: 12 Mar 06
Posts: 80
Credit: 53,305
RAC: 0
Message 7801 - Posted: 20 Feb 2008, 18:46:31 UTC - in response to Message 7800.  

I was writing this in an email, but it might be useful to others so I\'m putting it here instead. You said you were using a cube with the subsurf modifier\'s Render Levels setting at 6 to make it round. This generates A LOT of faces, which will require a lot of RAM and slow things down a lot.

A subsurfed cube has this many faces:
Level 0 => 6 (no subsurfing)
Level 1 => 24
Level 2 => 96
Level 3 => 384
Level 4 => 1,536
Level 5 => 6,144
Level 6 => 24,576

The number of faces increases exponentially, in this case it\'s 6*(4^L)

The reason for the sudden buildup of RAM is because the Render Levels setting is not applied until you begin to render your scene. This means that when you start rendering, each of your subsurfaced cubes suddenly goes from 6 faces to 24.5 thousand faces.

(Oh, and does ray tracing the reflections really require much more RAM? I wouldn\'t expect that... Oh well, either way RayMirror will definitely take more time.)


OK I think I\'ll try to reduce the polygons and the other stuff you wrote to me in your e-mail. The reason that it builds up a lot of RAM is apparently (as I understand you) that I have 20 million plus faces on the balls alone and some 80,000 faces on the cylinders alone :) Thanks for your advice. Actually on a side note. I tryed to make just 16 cubes. Also they appeared to build up a lot of RAM despite the fact that there should only be 96 faces overall. But anyway I\'ll try what you suggested in your e-mail to me, and see if it helps, and then I\'ll get back to you :)
ID: 7801 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
AC
Project donor
Avatar

Send message
Joined: 30 Sep 07
Posts: 121
Credit: 143,874
RAC: 0
Message 7802 - Posted: 20 Feb 2008, 20:30:28 UTC - in response to Message 7801.  

Try rendering a scene with nothing in it -- there\'s still a good bit of overhead required for a big HD render. I ran a test with everything I could think of disabled, and Blender still said it was using 40 MB, Windows reported a peak of 120 MB (including the program itself).

Actually, this raises a BURP-specific question: I\'m assuming that a good portion of that overhead is the storage needed for the image itself. The HD image is 2 megapixels, and a simple 32bpp RGBA image will need 8 MB. I\'d guess that Blender is using floating point values internally and just converting the output, so there may be an additional 20-40+ MB worth of internal storage including z-buffers and whatnot. If a client is rendering only a small section of that image (one of many parts) then will the amount of overhead be reduced accordingly? It doesn\'t matter for complex scenes because the objects will outweigh the overhead, but for simple scenes it could impact the minimum RAM requirement for BURP clients.
ID: 7802 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile Janus
Volunteer moderator
Project administrator
Avatar

Send message
Joined: 16 Jun 04
Posts: 4556
Credit: 2,097,282
RAC: 0
Message 7803 - Posted: 20 Feb 2008, 21:04:39 UTC - in response to Message 7802.  

Actually, this raises a BURP-specific question: I\'m assuming that a good portion of that overhead is the storage needed for the image itself. The HD image is 2 megapixels, and a simple 32bpp RGBA image will need 8 MB. I\'d guess that Blender is using floating point values internally and just converting the output, so there may be an additional 20-40+ MB worth of internal storage including z-buffers and whatnot. If a client is rendering only a small section of that image (one of many parts) then will the amount of overhead be reduced accordingly? It doesn\'t matter for complex scenes because the objects will outweigh the overhead, but for simple scenes it could impact the minimum RAM requirement for BURP clients.

Yes, the RAM requirements are typically smaller when using splitframe rendering. Although this only seems to be the case for a certain percentage of the sessions.
ID: 7803 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
KEP

Send message
Joined: 12 Mar 06
Posts: 80
Credit: 53,305
RAC: 0
Message 7804 - Posted: 20 Feb 2008, 22:08:05 UTC

@ AC and Janus:

Thanks for your advices on how to make these low render meshes. Now I\'ve finally managed to get my DNA whalestring down below 256 MB RAM according to the windows task manager. I really don\'t think that it will change much with the last few additions and the additional lights, so now everyone will actually have a chance to help out :) Also according to the test render I did, the frames will only take currently around 2 minutes, but that I think will increase as I put on more lights. But hey that will all be more obvious as I completes the last fideling. But here we have it, just as I was about to give up, it all the sudden succeded :)
ID: 7804 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
1 · 2 · Next

Message boards : General talk : Quite a few questions \"that need answers\" :)