Editorial Policy
   Contact Information
   Advertising Info
   Publicist Info
   Privacy Statement
   Contact 3Dgate

REVIEW • February 12, 2001

Affordable Graphics Cards for 3D CAD

by Peter K. Sheerin

Changing Cards

Whenever you are upgrading to a newer graphics card, you may be tempted to just shut the system down, swap cards and then install the new driver. Even though this is how things should work, doing so can lead to a difficult installation. Instead, restart the system in Safe Mode. (You may have to push "F8" just before or during the initial boot phase of Windows. Sometimes there is a prompt and sometimes there isn't-generally immediately following your system's BIOS start-up sequence. Make sure you don't choose "VGA mode"!) This starts up the computer with the plain-vanilla VGA driver built into Windows, enabling you to uninstall the (now unused) driver for your existing graphics card. For Windows 98 and 2000, this can be accomplished by launching the Device Manager, expanding the Display Adapters item and choosing "Uninstall..." from the context-menu (right click on the item). Next, shut the system down, install the new card and reboot. This should ensure that the old drivers are deactivated and removed from the system lest they interfere with the installation of the new card's drivers. This won't solve all problems (some drivers stubbornly refuse to be uninstalled completely), but it will go a long way towards making the installation of that new graphics card far easier.

As computer processors become ever faster, your existing workstations are likely to become slower than that new home computer your neighbor's kid got for the holidays to play the latest video game on. One of the most logical ways to upgrade your system's performance is by installing a new graphics card. In many cases, doing so provides a good value and a noticeable improvement, but sometimes the difference is only slight and hardly worth the effort. It all depends on what applications you use and which areas of those programs you care most about.

For example, if you're only doing 2D work with your systems, it has been clear for quite a while that there is almost no significant difference in performance between any shipping graphics card. Almost all improvement in 2D CAD performance today is coming from faster processors. Thus your decision as to whether to upgrade rests largely on any increased functionality that may be available, such as enhanced drivers or a digital-monitor connector.

There are many more differences in 3D performance, however, between different cards and between different applications-even different areas of a given application.

The test system for this review was our Dell Precision 410 workstation, with two 450MHz Pentium III processors, 384MB of RAM and a Diamond Viper V770D graphics card running Windows 2000. On this platform we tested two cards from ELSA, Inc., and one card each from ATI Technologies and 3Dlabs. The results are shown in Table 1 (note that the ATI card was unable to properly complete the 3DS tests, so I've omitted its scores). We had initially planned to test 3Dlab's latest card, the Wildcat II, but due to its use of AGP Pro, it does not work in our older workstation. Although AGP Pro cards are currently rare, this probably means this will be the last time we will use this trusty old system for graphics card benchmarking—it's clearly time to move to something faster and more modern.

3Dlabs VX1 is the oldest of the cards tested here, but still performs well and represents a good value.

The ATI Radeon graphics card is heavily targeted at the consumer market, with emphasis on DVD playback quality and performance and 3D performance in Direct3D-based games.

With the 3D Autodesk AutoCAD tests (run on AutoCAD 2000), I found dramatically different results among the cards. The ATI card, although newer than our reference Diamond Viper card by more than a year, turned in slower scores almost across the board. The 3Dlabs VX1 performed noticeably better in all but one test (the wireframe portion of the AutoCAD "57chevy" model). Both ELSA cards performed significantly better across the board, but both also boast higher prices than the others tested. While the three cards that showed improvement across the board did so moderately in the AutoCAD and SolidWorks tests, they all showed the most dramatic improvement in the Disreet 3D Studio MAX tests, and particularly those that emphasized manipulating large amounts of data, such as textures and complex lighting.

Missing from this test are the Autodesk Lightscape viewer tests. The program has not been updated in several years, and I've had problems getting it to work with a number of different graphics cards and platforms recently (including the ATI card in this test and several of the units from our recent laptop tests). Until I can determine the cause of these problems or replace the tests with a more modern architectural walkthrough product, the Lightscape viewer will remain excluded from future performance tests.

Driver Installation Woes

In conducting these tests, I had problems installing nearly every single card which made the whole experience frustrating (and far more time consuming than should have been necessary). These problems did not factor into my grading of the cards this time, but it will be a factor in the future. Aside from expecting easy installations that don't require the removal of other graphics card drivers, I will be looking for the posting of drivers not only on each vendor's respective Web site, but for the drivers to be included on Microsoft's Windows Update site, so that Windows can locate the appropriate drivers over the Web automatically, without requiring manual searching and downloading. The problems I encountered varied from an inability to install the driver for a new card—despite following the instructions included with the driver—to a strange condition in which physically installing a card and rebooting put Windows into a never-ending loop of rebooting the system every time it was partway through the startup sequence. I encountered all of these problems by trying to take the easy way out—just dropping in a new card and installing the drivers over those that were already there. The only way I was able to overcome these problems was by following the steps outlined in the sidebar "Changing Cards."

Consumer-Level Graphics Cards

A growing number of readers are asking my advice about the use of consumer-level graphics cards (which are usually much less expensive) as opposed to cards designed specifically for CAD users. The inclusion of the ATI Radeon was my first look at cards in this category. The results indicate that the ATI is not suitable as a replacement for a CAD-level OpenGL graphics card, but I'm not yet ready to write it off. Up until now all of my 3D tests have been based on OpenGL, with no effort given to test Direct3D performance. It may very well be that this card runs well with other CAD applications, perhaps using Direct 3D instead of OpenGL.

To test this hypothesis, I reconfigured 3D Studio MAX 3.1 (the only Direct 3D-capable program in our test suite) to use Direct 3D acceleration instead of OpenGL, and ran the 3DS benchmark again and repeated the same experiment with the fastest card that I tested here, the ELSA GLoria III. Although I was not able to get the ATI Radeon to complete the 3DS benchmark (due to a problem with the benchmark, not the card's compatibility with 3DS) in OpenGL mode, it ran flawlessly in Direct 3D mode. Its scores on the portions of the benchmark that did run under both configurations, though, were much slower in the Direct 3D configuration. I found about the same performance drop with the ELSA GLoria III card, indicating that Direct 3D is not the optimum configuration for running 3D Studio MAX 3.1. This is not necessarily the end of the story, however. Representatives at ATI say they will soon release an updated driver that improves the card's OpenGL performance, and among the many CAD applications that we don't use for benchmarking there may be several the card actually does perform well with. For instance, the merlin 3d modeling and visualization tool from Digital Immersion relies entirely on Direct 3D for its graphics acceleration instead of OpenGL and might perform better with the ATI Radeon than with cards focused on the OpenGL CAD market. I will be evaluating other programs and benchmarks to add to (or replace) our existing suite of tests, with an eye towards a more comprehensive representation of the software available for CAD and visualization.

ELSA Synergy III (the retail version of NVIDIA's card) offers a significant performance upgrade over many older cards.

The ELSA GLoria III is similar to the Synergy III, featuring the same set of connectors, the same size card and the same drivers—differing only in the performance characteristics of the graphics chip it uses.


An item related to the question of using consumer cards for serious CAD work is the amount of testing graphics card vendors put their cards through. Many CAD vendors (including Autodesk, PTC, Solid Edge, and SolidWorks) have certification programs designed to assist users in purchasing hardware that is known to be compatible with specific software. Each of these "logo programs," as they're called, relies on a test suite designed to make sure that any card bearing the logo will perform correctly—though not necessarily fast—with the application. But since this testing is time-consuming and often turns up problems in the drivers that need to be fixed, it is not often done on consumer-level cards (although ATI's older cards are certified for PTC's Pro/ENGINEER).

Even when a graphics card vendor takes the time to complete this process, I have found it difficult to determine if a specific card is certified for a particular version of a CAD application. I can sometimes find a list of applications on the graphics card's Web site, or a list of cards on the CAD application's site, but rarely does the list of applications include the version (leaving you to guess if the version you use is covered). Moreover, some of the lists I've found on the CAD vendors' sites are out-of-date, including mostly year-old cards in the list.

Future Expectations

Next time I look at graphics cards, I'll have a new set of criteria beyond just benchmark scores-expect me to kick it up a notch or two (can you say "Bam!"?) and look for the smoothness of installation, stereoscopic support (I'll be testing both built-in support and the VGA dongle retrofit stereo glasses), certification for CAD applications and compliance with such standards such as DVI and DDC2ci (the latter of which enables software control of monitors).

3Dlabs VX1

3Dlabs is one of the most established players in the OpenGL graphics card market and has the broadest range of cards designed for the CAD market. Once one of the leading 3D cards, the Oxygen VX1 now has a comfortable position as an entry-level 3D card for the CAD market and has a noticeable boost in performance compared to older generation cards such as the Diamond Viper in our base configuration (as can be seen in Table 1). It has 32MB of SDRAM, multi-threaded OpenGL and Direct3D drivers (enabling it to gain modest benefit from dual processors) and a useful set of software tools. For 2D AutoCAD users, the inclusion of Vibrant Graphics SoftEngine 4 provides a useful set of viewing tools and a performance boost in some cases. And E-Color Colorific software provides relatively easy color calibration without the need for any hardware calibration devices.


3Dlabs VX1 is the oldest of the cards tested here, but still performs well and represents a good value.

Pros: Lower cost; decent performance.
Cons: Older card design; no digital monitor support; no stereo sync connector.
Price: $299
3Dlabs, Inc.

ATI Radeon

The ATI Radeon graphics card is heavily targeted at the consumer market, with emphasis on DVD playback quality and performance and 3D performance in Direct3D-based games. With its included transform and lighting engine (T&L), 32MB of DDR RAM, full-scene anti-aliasing and many other quality improvements for textures, the Radeon is capable of very good 3D performance—but apparently not with OpenGL CAD applications, where it turned in scores slower than the original Diamond Viper V770D in our lab workstation. I'm disappointed with the results from this round of testing, but not ready to write off ATI's cards for the CAD market. Our next round of testing will include more cards in this segment of the market and more tests designed to determine if other CAD applications (or different configurations of our existing tests) can take better advantage of the graphics power these cards have.


Designed primarily for the consumer market, ATI's new Radeon graphics card doesn't perform well on any of our current 3D tests, but if you don't do much 3D work in CAD and want a card that you can have some fun with in your home office after all of the real work is done, it would be a good choice.

Pros: TV out connector; DVD software player.
Cons: Sub-par OpenGL 3D performance.
Price: $200

ELSA Synergy III/NVIDIA Quadro 2 MXR

ELSA and NVIDIA recently entered into a strategic partnership, with NVIDIA taking over all of the hardware and software design efforts for the workstation-level graphics-card market and splitting up the marketing and distribution of the cards, with ELSA selling into the retail channel and NVIDIA supplying OEM vendors directly. In every case the cards and drivers offered with the two cards are exactly the same. From the test results it is plain to see that the Synergy III performs well in all of our tests and represents a great value. The card features both an analog VGA output and the new DVI-I digital/analog connector, and an S/Video connector for television output. On a CAD-level card a stereo synch connector would have been preferred over the TV output, but some of you may find it useful for some presentations or as an easy way to capture animations on videotape.

With the new integration between ELSA and NVIDIA, the familiar ELSA display tools for AutoCAD were not available for either of these cards in time for this review, but both companies plan on integrating those tools into the driver shipped with the card shortly (and thus available to both retail and OEM customers). Overall, the Synergy III offers great performance with 64MB of RAM at a price just under $500.


ELSA Synergy III (the retail version of NVIDIA's card) offers a significant performance upgrade over many older cards.

Pros: Very good performance; both VGA and DVI-I connectors.
Cons: No stereo synch connector.
Price: $499
ELSA, Inc.

ELSA GLoria III/NVIDIA Quadro 2 Pro

The ELSA GLoria III is similar to the Synergy III, featuring the same set of connectors, the same size card and the same drivers—differing only in the performance characteristics of the graphics chip it uses. NVIDIA claims that the Quadro 2 Pro-based solution has twice the memory bandwidth of the Quadro 2 MXR used in the Synergy III. Although this advantage doesn't translate into twice the performance of the other NVIDIA-based card in the vast majority of our tests, the difference is noticeable or significant in most of them—and close to twice as fast in a few of the most intensive 3D Studio MAX tests (notably the Texture1 model).


Although the most expensive card here, ELSA GLoria III (and its NVIDIA OEM twin) offers the best performance for any card under $1,000 and does so in all of the 3D applications we tested it with. It's a great choice if you need as much 3D speed as possible without pulling more than 10 C-notes out of your wallet.

Pros: Great performance; both VGA and DVI-I connectors.
Cons: No stereo synch connector.
Price: $1,119
ELSA, Inc.

Peter K. Sheerin is senior technical editor of CADENCE. Reach him at [email protected].

(This article first appeared in the February 2001 issue of CADENCE.)