thanks to @Thanzex who found the benchmark and lapineige who fixed the benchmark to work with 2.7, here’s a viewport benchmark to have an idea of how graphic cards handle viewport.
CPU : i7 3930K @4.2GHz
GPU : GTX 580 3GB
Screen resolution : 1919 x 990
Overall score : 7.2
Overall Time : 2 min 27 sec
What surprises me is the Sculpt mode Matcap is twice faster than the Solid Mode (19.41 fps vs 10.94fps). I thought matcap was slower than OpenGL lights?
CPU : i7 Q 720 @1.60 GHz
GPU : AMD Mobility Radeon HD 5000 Series (1GB)
Screen resolution : 1599 x 820
Overall score : 16.0
Overall Time : 2 min 14 sec
Looks like AMD GPU wins in OpenGL. Would like to see a result from somebody with a professional Nvidia card like Quadro or Tesla, because these should perform well in OpenGL
I did not know there was a benchmark for the viewport, great!
Well, I knew that I have the worst nvidia GT 430 in the market (Biostar), but I did not expect my integrated intel HD4000 is much better. So, I’ll have to research about how to solve some problems that I have with intel (Blender closes with segmentation fault when selecting objects)
nvidia
OS: Kubuntu 14.04
CPU : i7 3770
GPU : GT 430 1GB - nvidia 331.38
Screen resolution : 1919 x 994
Overall score : 6.4
Overall Time : 3 min 6 sec
intel
OS: Kubuntu 14.04
CPU : i7 3770
GPU : iGPU HD4000
Screen resolution : 1919 x 994
Overall score : 10.7
Overall Time : 1 min 58 sec
Edit:
So, Are these benchmarks ok? Curious results in ATI vs nvidia on viewport, particularly those in ATI where they get high scores with no really good time. :eyebrowlift2:
Edit 2:
I solved the intel problem by restoring Blender settings.
CPU : i7 Q 720 @1.60 GHz
GPU : AMD Mobility Radeon HD 5000 Series (1GB)
Screen resolution : 1599 x 820
Overall score : 16.0
Overall Time : 2 min 14 sec
Looks like AMD GPU wins in OpenGL. Would like to see a result from somebody with a professional Nvidia card like Quadro or Tesla, because these should perform well in OpenGL
This is what I would like to see as well; along with some more AMD / FireGL cards.
on Nvidia cards the FPS is capped at 60fps by the driver, artificially lowering the results. For example, on my 7970 I get 517fps on the 50k faces Solid. Any Nvidia card gets ~60fps on this test.
in edit mode the CPU is taxed much more than the GPU, making it impossible to make useful comparisons in regards to GPU performance between systems.
the tests include models with active subsurf modifiers, which slow down Blender, and are not GPU optimized at all. They should be applied before testing.
some of us may have anti-aliasing activated in Blender, again skewing the results.
the draw method should be identical for everyone, but some of us use a different draw method while testing.
version 2.71 beta runs faster on average in this test than version 2.70. Especially in edit mode 2.71 seems faster.
there is no textured model mode test with bitmap textures at varying sizes. That would be a telling test. For real world work at least one textured semi-complex model should be included in a GPU test.
The 60fps cap alone completely invalidates any comparison between Nvidia and AMD cards.
The bolts scene, for example: when I add a subdivide modifier set to 3 (3,110,544 faces) runs at 2.72FPS. When applied it runs at a healthy 80fps on my system (without anti-aliasing). The subd modifier should be kept out of ANY GPU test in Blender.
All in all, this test is fundamentally flawed on so many levels it is not really worth our time. Which, I believe, I already stated when it was posted on these forums the first time. It tells us a lot about the caveats in Blender’s performance on different levels, but tells us nothing useful about the GPU performance on different systems.
I appreciate the effort, though. Someone should come up with a well-thought out one.
I think maybe it is as you say. With intel I get better scores than nvidia, but when normal blender usage the viewport works really slow with intel (linux). Furthermore, with the systems in similar conditions and with the same graphics card, sometimes I get different results when running the benchmarak.
I do not know…
Edit:
So, now I do not know if what I said above is correct, you see the two subsequent messages
Disabling VSync in nvidia-settings I win only a few seconds. But disabling double sided now I get:
Overall score : 11.8
Overall Time : 1 min 24 sec
Before I got:
Overall score : 6.4
Overall Time : 3 min 6 sec
So “double sided” was the culprit, regarding the difference I noticed when using blender with nvidia compared with intel in relation to the results I got before.