Professional 3D Accelerators in Solid Edge 10: August 2002

We used to examining operation of professional accelerators in DCC (Digital Content Creation) applications. But their range of application also includes a wide range of other types of programs called CAD. CAD applications are mostly meant for professional designers and engineers building models of objects; in short, such applications substitute usual whatman paper and a drawing board. For CAD applications the WIREFRAME mode of an accelerator is the most important; we examined this mode in the tests of DCC applications as well, but in CAD it plays the leading role. For example, if a drawing is untidy will the examiner accept it? That is why hardware line acceleration is so important in operation with such systems. The idea of this mode is to render simple lines. Our test will show how different accelerators cope with this task, because a lot of them support the hardware line acceleration including anti-aliasing, which lets users hope for high-quality line rendering without speed losses.

Quite often CAD applications are called Mechanical CAD, we will call them simply CAD. Almost all professional accelerators are certified for most popular CAD applications, which means that a given accelerator passed the test in a certain application and usage of a given accelerator is confirmed by a manufacturer of a given CAD package.

Today we use a CAD package called Solid Edge from Unigraphics. Unfortunately, its version is not the latest - 10. The package itself doesn't have any benchmarks for testing accelerators, but the experts from http://www.spec.org/ released a special suite exactly for this version, which can be downloaded from here after filling in a simple blank.

There is nothing much we can say about the test. It allows measuring performance of a system regarding CPU, Graphic, file system etc. But we are interested exactly in the video system performance. Apart from the Solid Edge packet this test requires installation of the Microsoft Office, removing marks from the disc and not touching the mouse. That is why, having turning the mouse head over heels we start the measurements. The test is based on operation of scripts. Choose a mode, and a certain script will start running; it will open a precalculated mode of an object and fulfill certain instructions, for example, rotation, twisting or scaling.

For more detailed information on the SPECaps project go here.

Testbed:

Intel Pentium 4 2200 MHz (l2=512Kb)
ASUStek P4T-E mainboard (on i850)
512MB RDRAM PC800
Quantum FB AS 20GB hard drive.
ViewSonic P810 and ViewSonic P817E (21") monitors.

The following professional cards were used in the tests:

As you can see, the most complete lines of video cards are ATI FireGL and Quadro of NVIDIA. Besides, I also included some gaming cards:

NVIDIA GeForce2 TI
NVIDIA GeForce3 128MB
NVIDIA GeForce4 MX440
NVIDIA GeForce4 MX460
NVIDIA GeForce4 Ti4600
ATI RADEON 7500
ATI RADEON 8500
ATI RADEON 9000
SoftQuadro DCC 128MB
SoftQuadro4 750
SoftQuadro4 900

The following drivers were installed:

The ATI RADEON family was tested with the 6118 version, the ATI FireGL2 - ver.2088, the ATI FireGL8x00 - ver.3036. The NVIDIA based cards were tested with the latest official available version - 3082. The 3Dlabs Wildcat VP870 ran under the ver. 0423. The VP870 was estimated both in the maximum geometry mode and in the maximum texture one.

Why have I included gaming cards into the tests? I got a lot of letters where you ask me to use them as well because they are much cheaper and more widespread.

The tests were carried out under the Windows XP Professional. The drivers (system ones and according to devices) and the Solid Edge 10 packet were installed on the virgin operating system. The Vsync mode was disabled, the resolution was set to 1280x1024x32bit (as the test required), and an optimization in the OpenGL for a given application, here it's Solid Edge, was enabled when possible. If it wasn't possible to use such optimization the OpenGL defaults were set. After that we installed the SPECaps packet with a batch of tests for the Solid Edge. Here we chose the one which measured the video system performance.

All cards seemed to fulfill the tasks identically. No artifacts were noticed, but the test required only rendering of lines (wireframe mode) and Solid Shaded, which is not a too complicated problem for modern accelerators.

The benchmark tests the accelerators in the following modes: Solid Shading, Wireframe and Solid Shading Textures. It measures time and fps a given scene needs for rendering. After that the test gives the score.

So, the first test is carried out in the Solid Shading mode. SE stands for optimization enabled in the OpenGL for Solid Edge. This optimization can be used on almost all professional cards. None of the cards, professional or game, made us complain of the rendering quality. All produced identical quality.

Now let's estimate the speed of rendering:

Well, while the professional FireGL 8800 of ATI outscores the gaming cards of this company, the NVIDIA's gaming GeForce 4 Ti 4600 card beats its professional solutions. The situation is intriguing. At the same time, the 3Dlabs Wildcat VP870 is an absolute leader, though the victory in the texture optimization mode looks strange. But the wireframe mode comes with a precalculated texture. That is why the victory exactly in this mode becomes correct. Besides, the gaming cards are able to work with the Solid Edge! Sometimes they even outperform the professional ones!

The second test is a pure wireframe mode. The hardware line acceleration has the most considerable effect here. It's interesting that the oldest professional card FireGL2 is leading here. It doesn't run far away from the other cards, but the breakthrough from the latest ATI's card is great. The NVIDIA's professionals perform according to the price and performance levels, only the Quadro 4 900 doesn't fit in here. The higher score of the Wildcat VP870 in the texture mode in comparison to the geometry mode looks odd, probably, because of the drivers. The gaming models show quite good results, except the pure outsiders.

And the last video test is SmoothShading Textures

The Wildcat VP870 takes the palm here. It is followed by the Quadro4 900 XGL, the most powerful NVIDIA's solution; the third position is taken by the SoftQuadro 4. The Geforce 4 ti 4600 looks much weaker than both Quadros; it means that it's possible to make an almost professional card out of a gaming one just by altering the drivers. Among the ATI's solutions the FireGL 8800 looks the best one; the gaming RADEON 8500 is close on the heels. The outsiders are the same.

Now, when we studied all the cards in separate tests, it's time to look at the whole picture:

Well, there is nothing surprising in this test. I just what to say that the professional cards cope much better with the tasks than the gaming solutions, that is why their price is justifiably higher. Nevertheless, you shouldn't treat all cards equally, even within a professional class. Depending on speeds and capabilities they have different price tags and different preferable areas of application. Our tests have revealed the leaders. But do remember that the Solid Edge is just one of a big army of the automatic designing systems.

Additional test

During our testing NVIDIA released a new driver version, that is why we had to retest all NVIDIA-based cards with this version which, according to NVIDIA, provides better performance in SPECviewperf which lets us hope for higher results in real applications. So, here is the last diagram.

The light blue bars show the results obtained with the previous version - 29.42, the dark blue ones - with 30.82. On the whole, the situation is worse. Some tests have improved the scores, but have you noticed which cards? Quadro DCC ? It has too few certified applications. Quadro 2 MXR? I don't think the new version was optimized exactly for this card. Quadro 4 750? The difference is negligible. I can say that the new drivers are worse than the old ones. Nevertheless, the overall difference between the versions is not great, though the ver.29.42 is preferable.

That's all about testing in the CAD systems. New products, new accelerators, new drivers will fill up our future reviews. And take into account that optimization of drivers affects much modern 3D acceleration and especially professional modeling.

Alexander Kondakov (mailto:kondalex@ixbt.com)

Write a comment below. No registration needed!