V6 Goal: Display Performance



What graphic card do you have?

  • All nVidia is nicer than AMD
  • nVidia has good driver overrides, so you can set your Rhino AA like this, but it can make Rhino unstable as it does take into account Rhino’s settings, so if Rhino AA = 0 makes a different image thatn if Rhino AA=2 or 8.

    (You should be able to decipher the Norwegian)

And here is vanilla settings and Rhino 8AA (with the Quadro)

(Right click each image and open in new tab, then swap back and forth to see any differences.)

Edit: I had to remove the info about Quadro being nicer than GeForce, as I could not see any difference between the two when I did a test. Also The GeForce AA on a massive 2D didn’t show any speed difference between AA sat to 0, 2, 4 or 8.

Edit 2: Rhino 6 WIP did the test in 4.7 seconds while Rhino 5 did it in 6.5.


Hi Holo
That’s a nice trick. I will give it a try. I use Nvidia’s cards.
No problem reading Norwegian as I am “partly” from Sweden.


I just tested the latest WIP on a realistic typical large-ish scene using my default custom display mode - shaded, no isocurves, thick edges.

File size: 1.31 GB
Number of visible objects: 8 130
Polygon count: 5 380 237 quadrilateral and 6 736 397 triangular

TestMaxSpeed results - Time to regen viewport 100 times:
RH6: 21.01 seconds.
RH5: 58.67 seconds.

That’s pretty good! :clap: :relaxed:

Of course, that will just make me put much more of the overall system in one scene - something that hasn’t been possible until now. I’ll will therefore probably not experience any speed improvements in RH6. :imp:


Wim - can you run exactly the same test, but in wireframe (no shading)?


Results of the exact same test in Wireframe mode:

RH6: 30.73 seconds.
RH5: 38.64 seconds.


Thanks Wim. So even in an all-wires scene, the WIP is showing a c.20% improvement. Impressive!


Since my problem are frame rates of a few fps (0.5…2fps) I would be glad if we would talk about high speed improvements and no improvement of a few percent only. Since years the GPU power is grown and not used. If my GPU show me 15% usage than I expect a lot of potential.
During my last project with a large train interior the problem was not the slow display only, also the responsibility of Rhino was low. So, for me it looks like we don’t need some enhancements, we need a big step forward, a radical change of the way like things are solved. We talk about it since years and it should be on the time now. Speed improvements of several hundred percent are needed.

(Steve Baer) #169

I have seen up to 10x speed improments which is probably an extreme case. This is all going to be model dependent and display mode dependent.


For me only before and after scores counts. A jump from 0.5fps to 50fps would be phantastic, but not from 20fps to 200fps. In your case - was it a problem model with a low fps?

(Steve Baer) #171

I typically only look at problem models and yes this particular model was completely unusable in V5. I also agree that there really isn’t much point in trying to optimize models that are already fast enough to use. Generally I feel that above 30fps, the model is very usable. 60 fps is ideal, and above that it really doesn’t matter.

(Kelvin Cheng) #172

Here’s the display performance comparison between Rhino 5 64-bit and the current Rhino WIP. To make the test results comparable, all tests have been done on the same machine.

CPU: Intel Core i7-6700K @4.0 GHz


Display Cards/Chips: Nvidia GeForce GTX 770, Nvidia GeForce GTX 970, Nvidia Quadro K620 and Intel HD 530

System: Windows 10 64-bit

Test model: FireEscapes_Caustic_Test.3dm (152MB with render meshes. It’s modified from one of @BrianJ’s models, thanks.)

Information reported by the PolygonCount command in both Rhino 5 and RhinoWIP:
There are 265768 quadrilateral polygons and 186908 triangular polygons in this model.
There would be 718444 total triangular polygons in this model after forced triangulation.
(The model was not re-meshed in Rhino WIP.)

Test: Use the TestMaxSpeed command with default options. Test three times in every display mode and use the average.


My comments:
High-end GeForce cards have similar performance with the low-end Quadro card in Rhino. The price of a Quadro K620 is about a half of a GTX 770 or 970. (I don’t have a middle-level or high-end Quadro to test.)

Upgrading from GTX 770 to 970 does not make sense if it’s only used for Rhino.

Intel HD 530 may compete with middle-level GeForce cards. (Just a guess. I don’t have a middle-level GeForce to test.)

Intel HD 530 has greater improvement on Technical, Artistic and Pen display modes in Rhino WIP.

Intel HD 530 is 4x faster with hardware acceleration turned on than turned off. It can be hardware accelerated.

I didn’t see obvious display problems with Intel HD 530 in the test. I’ll try to mainly use Intel HD 530.

(I’m not promoting Intel HD series by the way.)


Nice test results, but not here.

At my machine I can open my last complex models per latest WIP now. For example a train exterior, file size 950MB.


Rhino 5 77s
Rhino WIP 88s

Also the display quality of the WIP is less good than on Rhino 5.

Rhino 5

Rhino 6 WIP

(Kelvin Cheng) #174

Does the PolygonCount command give the same result for this model in Rhino 5 and Rhino WIP?


@KelvinC - That’s an excellent review and analysis. Thanks very much for taking the time to put it together. It’s the first time I’ve been able to see comparative gains (if any!) between high and low level GPU’s, for frame rates at least.

In terms of display/image quality, were you seeing much difference between the four cards?

(Steve Baer) #176

What display mode are you train model tests made in? It helps me to know if this is primarily a wire drawing issue or a shaded mesh issue.


For me basic display modes are interesting, no special effects. Most I use the shaded mode without special effects to arrange a scene. Additional I use the rendered display without shadows to inspect textures and colors. Seldom I use the wire mode only.

I didn’t check the poly count of the bike model before. From my tests my impression is that any typical graphic card related parameters doesn’t show a big impact to the frame rate. Antialiasing and poly count - both are no problems for a GTX780. I can extract the render mesh and hide the NURBS and Rhino is running like a dream at v5.

My train model seems to be slow because it’s to heavy, the GPU load is quite high. Interesting observation at my train model - the WIP is using nearly 100% of the GPU power, but the frame rate is slower than at v5.
Also interesting - I extracted the render mesh at both versions and let it run with the mesh only - WIP was slow like before (88s->86s), v5 is faster than before (77s->66s).

Back to the bike model, because we can test it here and at your machines. Polygon count checked - 3.600.000 at both versions. But … extractrendermesh and hide NURBS -> WIP slow like before (above 20s) and at v5 it is much faster (mesh only (wires 0, edges 1) 5s, NURBS without isocurves 14s). At v5 I get a 300% improvement by using meshes instead NURBS.

From my v5 tests I expected that the model will at v6 fast like at v5 in mesh mode (or faster since I get a max GPU load of 50% only). At v5 we can switch from NURBS without isocurves to Mesh with edges - nearly the same visual look - but 300% speed increase.
This experience let me use meshes at my complex models often - for example a train full of premeshed seats. It looks like at v6 this trick will not work anymore, so, at v6 my models will be much slower than at v5.

(Steve Baer) #178

On the OpenGL settings page in Rhino WIP, what is Rhino reporting in the “Video Hardware & Driver Information” section?

I want to make sure the WIP is properly getting OpenGL set up. When the WIP thinks it is running on an older OpenGL system, it degrades to alternate methods for drawing geometry.


Hi Steve, here the current state:


Thank you Micha for your tests, I have to agree 100%, we need to have better options to make huge scenes faster and i am not talking 20 % but really faster…

I really hope that we will see some more progression here. We are doing architectural visualisations and we see that the 3D content we use from 3rd party companies to furnish our scenes gets more and more heavy each year in regards of texture sizes and polygons.
Trees, cars, persons and alike add tremendous amounts of weight to the scene and often I find myself waiting for Rhino to react to my input, especially at the end of a project. We try to use octane proxies wherever possible but sometimes this is not the most flexible approach…

(Steve Baer) #181

Are you also only seeing 20% performance gains in shaded modes for your models? It seems that Micha is currently a special case and I am trying to figure out why he is seeing the lower numbers.