I don’t know how much can help the soc project.
I have tested it,it gives 10 times faster redrawing,but it has to handle in a good way what happens when memory card finishes(when I have tested it crashed when going outside graphic memory)
EDIT
The coder I think have fixed this bug,but I haven’t tried yet.
I just wanna clear one thing up. When i get to about 5 million polys, blender is unsuable, it takes about 4 seconds to calculate each frame in the 3d view. It took ages to get to 12 million polys, but im trying again to get more 
I think what would be quite nice is if Blender had a way to dynamically subdivide geometry based on viewport performance. For example, only load/draw dense polygons nearest the viewing position to maintain the usability.
There’s no point in allowing Blender to hang up a machine. This is where arbitrary limits come from for subdiv modifiers. IMO, it’s better to just let users set what they like but have the limit hard coded in the display code so it simply will not continue to draw if the interactivity goes below 5 FPS or something.
It could even be some sort of depth function that only displays vertices in edit mode within a certain range - I know clipping planes can do this but something that knows to at least draw up to a certain interactive limit. In object mode, the meshes can be switched with proxies.
When it comes to sculpting, only subdivide the meshes round the working area - I guess this is more like hierarchical subdiv surfaces but I mean closer to LOD-type drawing systems.
If you are working on a sculpt model, change all other high res models to basic meshes in the background. As you rotate the mesh, the density of the displayed polygons changes based on the distance from the mesh and the focal areas you are working on.
Obviously if you have a powerful enough machine, you’d be able to draw everything without any limits but it should mean that even the lowest end computers are at least able to open and possibly render complex scenes shared by others.
I can’t even get past 1/2 million without crashing 
Dual core ATI Radeon 1600.
The draw mode is important too. Displaying wireframes is slower than without. Try subdividing an object to 50,000, turning the draw mode to bounding box and just keep duplicating.
I have 4GB Ram and only displaying bounding boxes, I got a scene up to 13 million before it crashed - wouldn’t go over 3GB Ram allocated to Blender.
It would be good to only have the bounding box data in memory while working but it would probably mean a lot of swapping to a slow hard drive. Even if it was a special mode though, at least it would allow the possibility of working with insanely complex scenes on low end hardware. You’d have to watch the object count though.
Thanks Trevin. That’s great - I hadn’t thought of copying the attributes. Still, I still feels a bit off that one cannot set the Drawtype for all selected objects at once with one push on a button - but I guess this is the ‘path of Blender’ (very database-like).
Hi guys, i’m new and i have a question regarding performance and system specs… and thought i’d try here before opening a new topic. I’ve just started using Blender 2.5 for sculpting with the clay brush. Yesterday i did my first sculpt actually (i’ve been using blender for box modeling a couple of months now). However, my setup gets sluggish in multires levels 8 and 9, barely usable for sculpting (+/- 750.000 polys). I’ve seen youtube clips of guys getting really detailed results, much finer than i seem to be able to get, and it “seems” they don’t suffer this sluggish experience (though hard to tell in timelapse video’s, lol). I made this model today and had to cut off the back of the head because it was keeping me from doing more detail on the face.
Anyway, the question: What is most likely to improve my speed? CPU, RAM or GPU? I’m working on a Windows xp machine, with 2 GB ram, a core duo E7500 at 2.93 ghz and an onboard Intel GMA3100 GPU. Don’t tell me to switch to linux because it’s never gonna happen, lol.
Or is this all normal? I can understand i need to get the hang of it all and get more familiar with the tools in order to get beter results with less resources.
Thanks!
Not having a dedicated graphics card is always going to hurt you big time. Also more ram, but you’ll be limited anyway not having a 64bit operating system.
There may be more optimised versions of blender at graphicall.org
Richard
Ok, thanks. So a dedicated graphics card (i read somewhere ATI isn’t a good idea for Blender?) and more RAM. Would it improve the speed if i switched from XP to a clean W7 (64bit) install? Or would the fact that Windows 7 puts more strain on the hardware even out the fact that it is 64 bit?
Or how big would the difference be jumping from xp to linux? I could always consider dual booting, though i think it’s a drag. I just can’t afford giving up windows for all sorts of reasons.
PS: i am correct assuming i need a 32bit built since i’m running on XP though my CPU is 64bit, right?
Yes. 32bit OS = Only 32bit software.
i managed to have 100 mio polys, even in viewport. ok, then you can’t work anymore but it’s possible. 
can’t resist to post a screen:
Kay, but polys are smaller in Berlin so … that doesn’t really count! LOL
I think you can do quite better with your actual hardware. When I was testing the new Multires with the default cube I had the same problems: low poly count and slow performance.
The new Multires needs a base mesh with more detail than the base cube, a base mesh that only needs around 4 or 5 levels of subdivision to reach the poly count you want.
For instance if you want to reach 10.000.000 polys, try using a grid with 100x100 quads or any model with around 10.000 faces.
There is also great speed improvements when using “Orthogonal view” instead of Perspective.
Try the option “Fast navigation” in the left toolbar in Sculpt mode for more pleasant rotations, panning and zooming.
And to optimize memory, you can lower the number for Undo steps in Preferences (around 5 or 10) and the memory for Undo (20 to 50 MB). Also use the “Save External” option in Multires modifier, and set Preview Level in Multires to zero.
Try those to see how it works for you.
The thing about having a detailed base mesh has to do with the way Multires works. From what I understand, Blender cuts the mesh and sends each part (each face of the base mesh with all its Multires subdivisions) to memory, instead of the whole object, to speed up things and save memory. I think the cuts are done over the base mesh edges.
So if you make a model that needs less subdivisions then less faces are sent to memory at a time, thus work faster.
That’s what I understand, but you have more info here:
http://www.blender.org/development/release-logs/blender-250/sculpt-and-multiresolution/
Maybe in future this can be something automated, and Blender could for instance cut the model automatically in a subdivided stage of it.
For instance in 5 leves less than the one we are working on. So the “cutted parts” that are sent to memory are always only subdivided 5 levels.
Hey, thanks for the heads up. I’m definitely gonna try that out (i already subsurfaced the base cube once and cut away the back and applied it, and started sculpting with multires from there for the model i posted and i did notice a mild improvement). Although i have another question regarding your reply:
[
Also use the “Save External” option in Multires modifier, and set Preview Level in Multires to zero.
](http://www.blender.org/typo3temp/pics/51e873776d.jpg)
I don’t have that option in my version of 2.5!!! I noticed it in a couple of tutirials, and i even reinstalled it, but still nothing? Is there a different version i need to install or a plugin or something? My multires tab is a lot more simplistic and doesn’t give me those options.
I don’t have that option in my version of 2.5!!! I noticed it in a couple of tutirials, and i even reinstalled it, but still nothing? Is there a different version i need to install
Try the last SVN version at http://www.graphicall.org (I think right now it’s SVN 25507)
Do the same but with 5 levels of subdivision before applying, and you’ll notice a much greater improvement.
Thanks, the SVN from http://www.graphicall.org indeed fixed that problem. I’ll try the other things you mentioned in the morning. Thanks again.
PS: Been browsing through the forums some more, and i found people saying good things about the gtx260 (150 euro). Maybe i’ll be getting that. Anyone know if the price difference with the gts250 and/or the 9800gt (both around 110 euro) is worth it, or is the 9800gtx (160 euro) even better?
Is the new sculpt mode integrated into the SVN builds on Graphicall?
If you mean the clay brush, then yeah. If not, i have no idea what you’re talking about 
I can get sculpts up to 8th level on my Q6600/2GBRAM/nvidia 9400 GT. At about 3-4 million polygons, it becomes a bit sluggish, but no crashes. Looking forward for using 2.5 and its sculpt optimizations once it becomes more than interface beta-testing.