ObjReader Community
WIP => WIP => Topic started by: Patrice Terrier on April 25, 2018, 01:07:55 pm
-
These are the models waiting.
BTW, one of them had more than 100 millions indices, and still playing in ObjReader while using 99% of the GPU ;)
...
-
... one of them ...
The topmost one? ;)
... more than 100 millions indices ...
:o
In 32 bits, such a model would probably fail to load at all due to low memory at an early stage of geo re-triangulation. And it would probably take a pre-VBO ObjReader minutes to render a single frame on its artificially affined one-core-only CPU...
... and still playing in ObjReader ...
Might still fail on a cheaper video card with just 1GB of VRAM. Needs testing to prove or disprove my supposition though.
... 99% of the GPU ;)
This figure usually means that the total of geo data plus indices plus mipmapped textures plus compiled shaders exceeds the total available VRAM, and the GPU has to swap portions of it into conventional RAM.
How much VRAM is there on your video card BTW?
-
The big one is the Bugatti Chiron (the 2nd)
chiron.obj = 2.88 Gb
chiron.mtl = 183 Kb
with 6478 meshes, and 2146 materials...
The one i am working on, shall be a decimated version, however i was amazed to see that i could load and manipulate this monster inside of ObjReader64 with a FPS 24/25, in non-turbo rotation mode!
See the screen shot for further details.
I am using a GeForce GTX 980M: with a VRAM dedicated of 4096 Mb, and GDDR5
+ shared memory of 8161 Mb
and a CPU of 16 Gb.
-
Amazing! 34 million polies! :o :D 8)
Keep the screenshot intact. People won't believe we did it. :)
(Patrice, can I also have that 2.88GB big one in its original state before you have decimated it to some more reasonable measure? :) )
-
Done, thanks! :D
Looking at its perfect triangulation in wireframe, I think its original file geometry has been coded as NURBS and then converted to polies automatically when porting the file from .FBX to .OBJ.
Strictly speaking, the Wavefront Object format also allows the model geo to be designed and written using Bezier curves and NURBS to keep the file size low. But the process isn't entirely clear to me. So I think we'll stick to the good old poly meshes. Besides I don't see many NURBS OBJ files around, if at all. ;)
-
Down from 34 million polygons to 6.5.
After three weeks work, this model is at least close to full completion...
-
Outstanding, Patrice!
-
Here is the release candidate version of the Bugatti Chiron 2018.
The size of the archive is too big to be used as a single attachment :(
You have to download it from here
http://www.objreader.com/download/demo/Chiron.rar
if you know how to further reduce the size of the archive, or to split it into several modules, please let me know.
-
Thank you, Patrice!
If the app that you've used to generate this .RAR archive is a genuine WinRar, then it should have an option to create a series of split archive files of arbitrary and equal size. The initial chunk file will have the name Chiron.rar, the next Chiron.001, then Chiron.002, etc.
They can be uploaded as separate attachments and then downloaded side by side in a common folder. Now if you try to decompress the initial file that has the .RAR extension, all the others will also be decompressed automatically and concatenated into a copy of original file or folder that you compressed into such a split archive.
I don't remember the exact procedure or command line switches but I think you can look them up in the WinRar user manual or command line help.
-
The Lamborghini Centenario roadster has been added to the collection
http://www.objreader.com/index.php?topic=20.msg2788#msg2788