[an error occurred while processing this directive]

Houdini Master Benchmarking

Several months ago I got a letter from one of our readers where he asked me to extend a list of test applications for professional video cards. That was an interesting proposal and today you can see what I have come up to. You should consider the today's article as a trial balloon. Only together with you I can choose the testing technique which would suit both you and me.

Houdini Master

Houdini Master is a key product of Side Effects Software. It's not a single product - the company has a lot of various packets to offer but Houdini Master is a key packet for 3D modeling. You can use this packet for training (the freeware version has some limitations), and for professional activity. See the details on all products of the company here. The only thing I'd like to add is that the Houdini Master was used to create some effects of the X-MEN Movie.

Testing technique

To test one or another product one must know how to do it. Thanks to our readers who help me in it I actually know it. Exactly such collaboration can give us an ideal testing technique. But since not all our readers are gurus in 3D graphics, I'll describe every step I made.

In the long run the testing was split into two parts

  1. Speed parameters
  2. Quality parameters

Well, in the first case we will have a lot of diagrams and in the second case a lot of screenshots so that you can yourselves select the best.

I chose several demo scenes which put the maximum load on one or another function of the modern professional 3D accelerator. But as I found out, the Houdini much depends on the CPU clock and FSB bandwidth and hardly depends on a video card. That is why the scenes I singled out weren't the most difficult but their dependance on CPU was minimal. Also I had to try various rendering modes of one or another test scene. Below you can see the scenes used in measuring speeds and rendering modes so you can check my results.

Test scene file name Rendering mode in viewport window Example
align_boat wireframe

flagtwosided flatshaded

grass wireframe

magnetsbubbles vexshaded

springpushcloth smoothshaded

So, first we set up the operating system and all necessary add-ins and drivers. Then we installed the Houdini Master and started it. The first time you launch it it requires registeration. Then goes rebooting. After it we set the resolution to 1600x1200 @32-bit color and started the packet. The program window takes the whole screen but it's not the full-screen mode. That is why make sure to switch to it. Then we load a certain scene, set the rendering parameters for the viewport and start up the 'performance monitor'. Then we select required parameters in the popped up window and tick off "Enable output" so that we can see the results. Since the rendering mechanism is quite tricky you should start up animation and wait until the total number of rendered frames reaches 8,000-10,000 at least. Look how the Houdini implements rendering. Every action is recorded by the operator, and then the whole series is sent for execution. Thus, a simple curve turns into a small program the Houdini processes. That is exactly why such mechanism much depends on a CPU. When the number of frames processed equals 10,000 or over, you can write down the average fps.

The technique is not complicated, and everyone can repeat it. But far not everyone has such set of professional accelerators that we have in our lab. That is why I will readily share the results we obtained.


I used all the professional cards we had and several gaming ones. I selected those gaming cards which I think will have the most effect on the industry development in the near future

Cards tested:

Test platform

The tests were carried out under the Windows XP Professional, Service Pack 1, DirectX 9.0a, with all required add-ins and drivers installed. Vertical sync, anti-aliasing and anisotropic filtering were forcedly enabled in the drivers. The driver versions used:

For the NVIDIA based cards - 45.23
For the cards on ATI FireGL 8x00 - 3056
For the cards on ATI FireGL X1 - 1039
For ATI based gaming cards - 6378
For 3Dlabs based cards - 3010535


Let me draw your attention to some interesting facts. The NVIDIA QuadroFX 1000 looks very nimble, or in other words, the NVIDIA QuadroFX 2000 works too slowly. I contacted the guys at NVIDIA to find out the cause, and their answer will be given at the end. I assumed that they would suggest using another version of the drivers or test packet. But I was wrong. ;-)
The gaming cards from ATI perfectly demonstrate that they are designed for games only. The scores are too low, though these gaming solutions serve a base for professional cards which look much better! The company clearly realizes the market separation and separately develops drivers needed for one or another market sector.

That's not all. Since the Houdini is strongly dependent on the CPU we shouldn't stop at this stage, and now we proceed to the qualitative tests.

Qualitative testing

This time we don't need the performance monitor, the viewport window must be rolled out to the full-screen mode. Then we load a scene, set a required rendering mode, find a certain animation frame and make a screenshot. The rendering mode was VEX WIRE SHADED as it loads the video subsystem most of all. The test scene was provided by my namesake @Hawk, Alexander, thanks you very much! The test results are shown below in the table.

Card Link to the test results

NVIDIA Quadro2 Pro


NVIDIA Quadro4 550XGL

NVIDIA Quadro4 980XGL

NVIDIA QuadroFX 1000

NVIDIA QuadroFX 2000

NVIDIA GeForceFX 5600 Ultra

NVIDIA GeForceFX 5900 Ultra

ATI FireGL 8700

ATI FireGL 8800

ATI FireGL X1 128MB

ATI Radeon 9600 Pro

ATI Radeon 9800 Pro

3Dlabs Wildcat VP870

There are almost no equal results. I must say that the 3Dlabs has too many errors on the Wildcat. As to the other cases, you can make your own conclusion.


So, thanks to our readers I added the Houdini packet to our test suite used for professional cards. Unfortunately, its speed parameters depend on the central processor  to a greater extent. Nevertheless, we got an excellent possibility to estimate qualitative parameters since quality of drivers is so much discussed these days...


Finally we got the response (three weeks later) from NVIDIA. Let's me quote it: However if I run FX2000 using VEX Shading (Which is what they indicate they are using) and I run FX1000 using smooth shading, then I obtain results similar to those that ixbt is seeing. It is possible that they are mistakenly recording results for the FX1000 using smooth shading rather than VEX Shading. Actually, I got no clear answer. Since I myself developed the testing technique and then tested the cards I affirm that all the cards were tested in the same conditions and with identical settings. And we publish the materials as is according to the scores obtained on the certified drivers. Our readers definitely have something to think about...

Alexander Kondakov (kondalex@ixbt.com)
[an error occurred while processing this directive]