I bought an nvidia GTX 260 last week and i’ve been kind of dissapointed, to say the least. The 3D viewport becomes extremely unbearable if I even think about makinga scene that looks somewhat realistic (so in otherwords, all the time). I can still return it for a full refund and I’ve been looking at the radeon 6870— which is $70 more than what I spent on the GTX. According to benchmarks, the 6870 is a significant improvement over the 260. The question is, how does it perform in blender? To me, benchmarks are useless— I want real world application, specifically, blender. My computer was basically built solely for blender, so this is all that really makes a difference to me. Does anyone have the 6870, and could you recommend it?
Here are the benchmarks I was basing my assumption off of
Yo get the 6870 ati works good with blender now and that card is way better.
If your ‘realistic’ scenes are million+ polygon scenes then it will be slow on a lot of cards.
Post a wireframe of your models with some specifications (texture size, number of lights, scripts running etcetera), and then people can tell you if it’s worth the upgrade.
I’m currently running a GTX 260 and I have very little problems with high-poly counts in Blender, so either your system is set-up wrong or you’re trying to push too many polygons, textures etcetera through the engine.
Pretty much, this is a tree from the nature academy. I’m getting about 3 fps when I am scrolling around, not to mention about a second of a delay. I have to turn off all subsurface modifiers to get about 15fps
I noticed it says 1.9 million vertices
I don’t even use textured mode in the viewport, but fwiw I use 1 tiled texture for the tree bark only
To answer your question the ATI card will definately outperform your current card and I know from experience that the newer ATI cards work fine with Blender. However you are viewing a ton of verts in wireframe mode you kinda have to expect it to be slow… Turning on optimized subdivide and solid mode will help a bit.
EDIT: Just wanted to add, I can sculpt on models with more than 4x the polys that you have there and still maintain 30fps with my card (with the proper optimization set and in solid mode), which if you check my sig you can see its 1 generation before the one you are currently looking at (and slower I might add).
That scene will be slow even on the ATI card you mentioned.
Blender isn’t particularly good at handling large amounts of polygons, with the exception being in sculpt mode which is highly optomized for large polycounts.
It’s always bad to have sub-division surfacing enabled for the viewport, unless it’s one or two pretty simple objects.
If you turn off all of the sub-division, just go to the scene sub-panel in the properties panel and turn on simplify and set it to 0 for sub-division you’ll probably get a large speed increase.
Also when working on objects like that in a scene you should always be making simplified proxy objects, just a basic mesh that follows the overall outline/ shape of the tree to take it’s place while animating or modelling other items, and then only for rendering use the high-poly mesh. Otherwise you’ll just end up with scenes that have a frames per second of less than one and you’ll not be able to do anything, regardless of what hardware you’re running Blender on.
My advice however is to upgrade the card as the GTX is quite an old card. But don’t expect drastic frame rate changes, perhaps 2 or 3 fps at best I’d imagine.
I forgot what it’s called, but it’s used quite often in online 3d apps, and computer games. adaptive geometry, the further away you can represent the geometry with fewer vertices/polygons. and the closer you get you can add more. There’s of course different display modes for each mesh object, but to have something automagically adapting you’re mesh display in the 3d viewport would be neat. Could of course have a threshold in settings so it’s only active when n-number of vertices are in the scene.
On my machine (i7@3,6ghz, ati 5870 1gb, 12gb ram, Windows Vista 64bit) the following scene works smooth:
That’s about 4 million vertices and 2.6 million polys. The trees I made in Cinema4d v8 and xfrog4 (which came free some time ago with 3dWorld magazine 
)
So, as far as the scene goes, at least twice the amount of your scene. All is smooth, except for one thing: selecting an object takes about a second/two seconds. That’s the only thing driving me crazy: I can go much higher than these poly numbers, but selecting an object in the viewport can take up to six seconds. Mind, this can be circumvented by using the outliner, which selects objects without any lag.
If someone could tell me how to solve this little problem, I would be eternally grateful!
PS You did turn on the VBOs in the system preferences tab, didn’t you? If VBOs are turned off, I get much worse framerates: about ten times worse.
I think from the performance comparisons I looked at, even my 5770 is faster than the 260, it’s been a great card so far too.
@ Herbert123: Thanks sooo much for mentioning VBO’s, I never knew there was such an option! I went ahead and subdived a different model I had with VBO’s enabled, and @ 3million vertices it’s pretty bearable. Not perfect, but 15-20 fps which is a HUGE improvement. I’m still going back and fourth between the two cards though, whether or not it’s worth the $70. But, again, VBO’s worked magic for me. It definitely makes me feel like I couldn’t go wrong either way, now that the GTX 260 seems to be doing alright. Thanks again!
heh thanxx for VBO tip 
didnt knew about that