I'd like to proclame a new record of number of polygons ever rendered in Blender: 1.15 billion tris with blurry reflections. Rendertime 23 min, Mem 200 mb
Can anyone top that?
I'd like to proclame a new record of number of polygons ever rendered in Blender: 1.15 billion tris with blurry reflections. Rendertime 23 min, Mem 200 mb
Can anyone top that?
36 Comments
Got RenderTime Records?
But does it count with an array?
Good luck blender. U rock.
200 Mb of memory only??I guess you have like an i7 core?
Yeah... I saw that on BA. The new render engine code seems to be sweeeeet. :)
Now I'm about what would the other apps do when 2.5 is out? :D
need that build!!! :P
hi… maybe i missed something!? is this part of the raytrace optimizations from the google summer of code project? Very interesting indeed. I am really looking forward to see blenders "new" raytracer.
ooops...I read on and found out.
with my newly gained knowledge: wow.
The poster missed something rather important facts. First this is an instanced object. So it's basically the same object duplicated over and over again. This is why memory is so low. Blender will still have issues once you hit higher vertex counts. This is not 1.5bil faces. It's a single object, and that data is just cloned in real time. Think about it for a bit:
1,500,000,000 faces * 3 points per face * 3 co-ordinates per face * 4 bytes per co-ordinate (32 bit floats) = 54GB. Now granted faces share vertices, but we should see way higher memory usage with a real-world scene.
Secondly, this scene is very regular. The new ray tracing algorithms employ some bounding volume and space partitioning schemes that are very fast for non random data.
So yeah, it's not extremely impressive.
Timothy, I completely agree with you. testing Blender in a Dell T5400 topped at 16 Gb of Ram, ( Malloc crash) The machine itself uses 32 GB, then Blender have more room to grow, but the limit of Blender Itself appears to be 16 GB. (FC10, Ubuntu).
The instanced objects uses Ram only when rendering, then, the real thing is to know what is the real usage during Render. Not the usage of the Scene itself. Better than this, if we use to trow that instanced objects in another layer, there will be no laggie.
I am expecting some things for 2.5:
- Caustics + Dispersion.
- More Ram Window for 64 Bit systems.
- At least 64 Threads. ( Next Mac= 24 Threads, Next supermicro four sockets=48 or 64 threads, beowulf cluster= OMG!!)
"Can anyone top that?"
...No problem... Modo 401 make 1.6 Billion Polygons with activated GI in 12.1 seconds and have a realtime preview rendering system. Show the bottom Video on the follow Link:
http://www.luxology.com/modo/401.replicators/
Show it to marvel... Have a nice Day :)
dd
@Timothy Baldridge you make a fair point. But, forgive me if I am wrong here, 2.49 does not handle memory instancing well. This same scene rendered with 2.49 will use more memory and take much longer to render if it renders at all. So it is a valid test of to see if Blender is handling instancing better.
I think its still early days to see how much of an improvement there will be when rendering real faces as opposed to instanced ones but it is a promising start.
@ddiethelm2:
That's really no problem!!
With Maya you can handle about 900.000.000 but with FG/GI!!!! (don't know if Blender-User know what that means)
Sometimes I think here is something like Kindergarden!
P.S.: 1.15 billion? Sounds much?! When it's the American description it's only 1.5 Mrd !!!
@Lollipop Maya might handle that amount nicely, but It's a joke of a 3d package if you compare it overall with blender.
@Dread Knight. "It is better to keep silent and be thought a fool than to speak and remove all doubt"
I love Blender a lot, but I understand what it is, and it currently doesn't have the ability to make Maya "a joke of a 3d package". Although Maya has my least favorite interface, it is considered one of the most complete and capable DCC apps around. Blender is still a toddler, it has a lot of growing to do.
Question: what is Mrd; 766,666,666.66 tris?
Acronym?: "Maya resolution display"
that's as much as I could try to infer from your comment...
@Lollipop, please enlighten me about this terminology :)
haha, this is pretty funny....
we'll be laughing about this "Record" in a few months, old benchmarks always make me smile.
@Lollipop
The question was: "Can anyone top that?" ... and I said: "Yes, Modo can do." Had he asked: "Can your Blender this," my answer would be the one ;)
Do not misunderstand me, I use Blender for years because many functions are very good. Only the rendering speed is very slow. But this is fixed in the 2.5 ;)
Was the test done without layers? I bet I could beat that record easily...
Add an array to make it a 3D grid of objects...
wait wait!!!!! I want to count to proove it! 1 2 3 4 5 6 7 8 9 10 11 12.... urm ok bear with me here... 13 14 15 16 17 18 19 20 21..... ok.... urm.... just a moment.... 22 23 24 25 26 27....... urm..... where was I?.... ok..... urm..... damn it...... 1 2 3 4 5 ..... urm...... ah...... :)
keep counting ....
BTW on what hardware is this done?
That's quite impressive. Can we only expect this optimization to be noticeable for gigantic, complicated scenes like this, or will this be worthwhile for smaller raytraced renders as well?
I think you guys are missing the point - of course it's instancing and of course other renderers have been able to do this for ages. It's impressive because this is very new for Blender, and you'd have no hope of doing this before Andre's most excellent GSoC project.
My comment to the statement of Matt:
Hear ye! You are all CG-nopes, aren't ya?!!
By the way: what in the US is 1 Billion is in Europe 1 Milliard!
But Billion sounds more!! ;-))
Matt,
That´s the point, the article that was wrote does not mention Gsoc or Andre´s work. At first sight, the article appears to be written to cause an great impact. But without the proper information, the article causes sparks between some friends, and doubts. We need more info about the hardware, specs, memory usage under render, etc, etc. And, of course, info about the project and how it really works.
Lollipop, I don't get your logic :)
...we will see how this handles in a production environment.
--
Kevin
Ivan: You're right, the article is very light on details. It would be good to clarify that it's a test of the GSoC project here:
http://wiki.blender.org/index.php/User:Jaguarandi/SummerOfCode2009
that must be a strong computer
Thats very impressive indeed. There are going to be come very nice blender forests and crowd scenes soon :)
I cant imagine why people are knocking this because it uses instances. Instancing is EXTREMELY powerfull
a couple years ago was impressed by being able to handle near 4 million faces in a single mesh in blender. the point for being impressed with that feat, and with the feats of rendering even instanced meshes, is that this is an OpenSource project that is matching (if not exceeding) the power of crazy expensive software.
be proud for how far blender has come. i've been a user since 1.8, and the things that you can do in blender now were unimaginable back then, except maybe for ton ;)
"Can anyone top that?"
…No problem… Modo 401 make 1.6 Billion Polygons with activated GI in 12.1 seconds and have a realtime preview rendering system. Show the bottom Video on the follow Link:
http://www.luxology.com/modo/401.replicators/
Show it to marvel… Have a nice Day :)
dd"
ya, but..... IS Modo free and Open source?
marvel at the open source.
I don't really think that it's cheating to use arrays... because theoretically if you had a large model you could just subsurf it and use the awesome sculpting tools to achieve higher polys.
Which could also be considered cheating.
CG is all about cheating. Being able to cheat on this scale is epic :)
@~Dusty : Lol :)
There should be a defo render benchmark for all the os versions of blender, and some form of cinebench farm test as well, yea, that'd be great!!!