Hi to all,
I found viewport performance in blender is very poor, with dense mesh lag and navigation is difficult (also selection is very slow)
I have a GTX460 1gb it work well like with other software.
A quadro or a firegl can perform better ?
If, like others users said me, quadro and firegl do not improve so much performance, I think is a problem in data flow management. IMHO, a more robust viewport, and better flux data management, will have benefit not only in modelling but in general in all blender department.
IMHO, I think this must be a priority (especially for a software who want be a complete suite)
Well, as much as I hate complaining, you may be interested to know there are two separate GSOC projects that are likely to at least partially address the issue.
I have wished for better viewport performance for a while too, it’s got something to do with the way Blender draws the 3D view that is made to run on older (read: ancient) hardware. It needs to be re-written so that we can take advantage on the sort of computer setups we have today IMO.
There is an issue with the Fermi (includes the GTX460) GPUs where NVIDIA has disabled hardware-acceleration for double-sided lighting in the consumer cards. What you can do is disable “Double Sided” under “Object Data->Normals” for every mesh object. It might also be the case that picking works even worse in Fermi, but it doesn’t really work that well anyway.
Independent of that (applies to all modern GPUs) you should enable VBOs under System->OpenGL in preferences.
A quadro or a firegl can perform better ?
Older geforce gpus, AMD gpus, and quadro gpus are unaffected by the aforementioned double-sided issue, but apart from that they are probably not much faster.
Thanks to all for the reply. I just did the setting suggested here, but the perfomance not gain so much.
As I said, IMHO, a strong viewport can help blender in all its departments (polymodelling, sculpt, animation ect.), so I’m happy to know there is two projects about
Thanks MadNinstrel, please, can you indicate this two GSOC projects?
I noticed once campbell explaining adding more 3d viewport modes will drastically add render time for the 3d viewport. It was something like first the wireframe are drawn, then ontop the solids are drawn etc. basically it has to render each view mode and add it ontop of each other or something in that way.
having modes just as flags or toggles on/off and draw them in a buffer then just send the single composite frame to the 3d viewport could be a solution.
then it doesn’t matter if there’s 2-3 or X numbers of viewport modes they’re combined in a image buffer and just a single combined mode is rendered.
I’m not familiar with OpenGL code but in other script languages we often have a display list, and it’s always better to do the effects in memory then combine to a bitmap and blit it out.
Sounds interesting, the viewport isn’t too shabby atm, but it certainly has room for improvement.
I wonder if the viewport uses object instances to their fullest potential as it is.
Off-topicish: Does the viewport support custom shaders in .glsl format like max supports realtime HLSL shaders in their materials? I kinda miss that in blender, maybe I haven’t found it yet. But i’d love to be able to use animated water shaders & such. You’d be amazed how nice you can make a realtime scene look with the right shaders & interface.
in the sunday meeting notes, I saw that dinges it gonna check into Cuda SDK 4.2 which means support for the new 6XX nvidia cards. Would be sweet to see some cycles benchmarks on those.
While waiting for some new code, a trick when you have a scene that is too heavy for the viewport to allow you to rotateor zoom around it smoothly while in Solid display, is to press press Z to go to Wireframe display.
Despite there’s no backface occlusion in Wireframe display, it will provide a big boost to the viewport performance, allowing you to rotate or zoom much more smoothly around the scene.
Then press Z to go back to your Solid display when you’re finished rotating.
