Test: 7 PCI Express 3D Pro graphic cards - BeHardware
>> Graphics cards

Written by Damien Triolet

Published on March 25, 2005

URL: http://www.behardware.com/art/lire/560/

Page 1


For many years now, ATI and NVIDIA have released all general public GPU in professional versions. These releases have tended to overshadow the specialized solutions. NVIDIA became the undisputed leader in this market, followed from afar by its competitor ATI. 3DLabs and SGI are also survivors of this industry and continue to release solutions for professional 3D. However, the market shares of these solutions are constantly on the downturn. How long can they continue to exist?

Not so long go, the requirement for a professional graphic card was to process as much polygon as possible. Requirements are changing however. Recent applications have been constantly improving the resulting quality, and have begun using pixel shaders. But SGI doesn’t offer any graphic card capable of processing pixel shaders and 3DLabs appears to have performance issues on this level. Unfortunately, we were unable to verify this as we couldn’t get hold of the last Wildcat from 3DLabs. There is one important factor in this transition, SGI now uses ATI’s GPU in several of their computers.

Development costs for recent graphic processors are so high that it has now become less and less profitable to develop a processor for a single niche market, such as the professional market. ATI and NVIDIA, which have a high level of productivity for the general public, have therefore decided to offer the same chips to professionals. This new offer has considerably reduced “professional” graphic costs. Of course, in time, some necessary optimizations, specific to professional applications, have been made. This is why ATI and NVIDIA use different denominations for the general public (Radeon and GeForce) and for the professionals (FireGL and Quadro): the aim is to clearly underline the two products´ basic differences. Could this be seen as another way of increasing the invoice?

The price
The first and easiest difference to spot is the price of the professional version of a GPU. ATI and NVIDIA have considerably reduced professional graphic accelerator prices, but still make more profit out of these products than with their other solutions. There are three reasons for this price gap:

Professional graphic cards are used in very expensive workstations (several thousand to several tens of thousands of dollars). Profits generated by these workstations are naturally much higher. And so, it is perfectly logical that ATI and NVIDIA want to adapt their profits. Several options and optimizations have been made and can only be activated for the FireGL and Quadro, in order to protect this market. These users also respond to productivity gains rather than quality/price ratios.

The other explanation is that professional graphic cards require additional developments for drivers, whether for the integration of several functions designed for professional applications, or the optimization of the same applications. This development has a price and this is only applied to the professional graphic cards.

Finally, the graphic card market requires a very high level of reliability whether for display purposes, or for the actual graphic card function. This implies more advanced validation tests and higher security margins than for general public products, as well as efficient and fast support in case of hardware or software problems. ATI and NVIDIA provide this type of support. NVIDIA maintains that they provide a better service but ATI vigorously denies it. Unfortunately it isn’t possible to cast a vote in favor of ATI or NVIDIA at this point.

Page 2
Drivers, OpenGL, upsides

The Quadro and FireGL drivers differ from the GeForce and Radeon even when they are based on identical files. One part of the driver’s code is aimed only at professional graphic cards and thus they outlaw general public graphic cards. Most of the 3D-creation software overloads the graphic cards with geometrical calculations; texture filtering and shaders are relegated to the background. The situation is the other way around for video games. Naturally, this is the case for most situations - at times the situation is different. Basically, however, it allows ATI and NVIDIA to optimize their drivers for this type of situation.

Another characteristic of professional graphic cards is how they cope with three-dimensional displays. This process is often very different from video games. The processor plays a much more important part, and most of the time the result isn’t completely processed through pipelines. That means that the second image only starts to be built when the first one is quite finished. So, with complex scenes, the processor and graphic cards often restrict performances, no matter how powerful the other factors. Reducing the processor load is very important even when the GPU is pushed to its limits. If one part of the CPU load is assigned through application, the other part comes from the API and the driver. ATI and NVIDIA are able to optimize this part of the process via their drivers. So that increases the driver´s importance even further, and can thus tip the balance in favor of one or the other according to its efficiency.

Each of them also have optimizations designed for several applications, as well as a specific profile which correctly parameters the driver.

OpenGL is by far the most common 3D API in the professional world. NVIDIA has developed a very good and acknowledged OpenGL driver, unlike ATI who has been attempting to catch up for many years. Fortunately for ATI, the driver quality has improved but NVIDIA remains the unchallenged reference in this area. It is important to know that, in the beginning, ATI’s professional OpenGL driver was developed separately in Germany and that it was fused with the general public driver (also under construction) several months ago. As a result of this fusion, the offices have relocated to California; where the excellent and acclaimed Direct3D driver is developed. However, this transfer, even if Ben Bar-Haim, VP of ATI Software, denied it, created many wide sweeping changes in the development team, since only a few developers from the original team have now joined the new one. It is obvious that the new team, even though it has been getting help from several developers of the original FireGL, was not totally efficient from the beginning when faced with this unknown FireGL code. During a visit to ATI’s Californian development center, Joe Chien, in charge of the driver development, told us has that it fully intends to increase the OpenGL driver´s quality to that of the Direct3D´s level, and that several improvement have already been made, and also added that it will be available during the next few months. So this test will provide an opportunity to see the improvements in ATI’s driver compared to NVIDIA.

We have also checked out the current support of high level shader programming language in OpenGL, the GLSL. A 3DLabs utility checks the correct functioning of all points, as well as the impossibility of using GLSL functions in unexpected ways, for instance by using a function with an incorrect parameter. If ATI obtains 98%, NVIDIA only reaches 48%, as most of the tests have been unsuccessful. We have asked NVIDIA to comment and their answer was that the 3DLabs utility is too strict. Nonetheless, it follows the specifications set out by the ARB OpenGL and it is most likely NVIDIA’s driver, which is too flexible and allows operations that do not correspond to specifications. What are the consequences? Chiefly, that there is no guarantee that GLSL shaders developed with NVIDIA’s drivers will work with other GLSL compatible drivers (this, however, makes the point for standardized languages…).

Anyway, technological demonstrations using GLSL shaders work perfectly for ATI and NVIDIA. It is also important to know, that most of the 3DLabs demonstrations are using shaders that are too complex for ATI´s current GPU and are therefore rendered in software.

It is also important to remind you that NVIDIA is using its own high level programming language (it is however possible to compile the shaders to the OpenGL ASM standard model), Cg. This language is very well introduced in the professional world and provides a great strategic advantage for NVIDIA because it is used in several applications.

Professional graphic cards advantages
Except for performance optimizations, these drivers also include additional functions.

The most important concerns points and lines anti-aliasing hardware acceleration. It differs slightly from anti aliasing used in games and that isn’t activated with the GeForce and Radeon. The graphic cards´ performances are reduced when it is activated.

The quad-buffered stereo mode provides a possibility, as is the case with a few general public graphic cards, to emphasize depth in a picture. Of course, it is essential to have stereo goggles or an advanced monitor. The object is to display images with a slightly different point of view alternatively, in order to stimulate each eye´s vision. Goggles for example, mask images for the left eye when they are intended for the right one and the other way around. As the name indicates, this mode uses 4 buffers instead of the usual 2 buffers (back buffer and front buffer), it also uses more memory, and needs more power to keep the same number of images “seen” per second compared to the standard display. The image must be calculated twice as fast.

Another major difference the software management displays are more advanced: first on the memory level as well as by the reducing of a certain amount of unnecessary calculations. For example, it is possible to display a menu (via the overlay planes) on a 3D display without touching the content of this 3D display and without needing to recalculate the image when the menu disappears. Other similar functions exist, but all of this is only useful in very specific cases, as, on the whole, the recalculating of that part of the image temporarily hidden by the menu, does not create problems.

Page 3
Graphic cards tested

For this test we have restricted our choice to the last generation PCI-Express graphic cards:

FireGL V5000

FireGL V5100

FireGL V7100

ATI’s graphic cards use R423 or RV410 GPUs. These GPU also support standard functions like their general public version: 2.0+ Pixel and Vertex Shader. Pixel Shaders are restricted to 512 instructions and don’t deal with branching and loops. If a more complex Pixel Shader is sent to the GPU with a professional application, in OpenGL, the CPU will calculate it and it will considerably reduce its performances. This is not, however, the case for current applications that are just starting to use shaders for real time display. The internal calculation format is 24 bits in floating point (s16e7), so far this format has been sufficient even for professional applications.

All these graphic cards are Dual Link and stereo mode compatible. With the dual link it is possible to connect a high-resolution monitor such as the 30” Apple Cinema HD Display (2560x1600).

We have had an unusual problem with the FireGL V5000 (this problem also exists with the Radeon X700 Pro). The power supply stage uses low costs components that vibrate or sizzle, according to the load. This noise is very disturbing because it isn’t regular and varies directly with what is displayed on the monitor. One could almost play a symphony, just by adding or removing objects in 3D Studio Max. Of course in a very noisy work station, this sound will go unnoticed, but this downside, even if it is only mildly irritating when dealing with a US $200 general public graphic card can be far more upsetting with a US $700 professional one….

Quadro FX 540

Quadro FX 1400

Quadro FX 3400

Quadro FX 4400

All Quadro PCI Express tested here are based on the 6th generation general public GPU. They support class 3 Pixel and Vertex Shader and are capable of processing very complex shaders in hardware with branching and loops. If this advantage isn’t yet available in practice, professional application developers could use them to write their shaders in a very natural way, like the rest of the code, without paying attention to hardware specificity. NVIDIA would thus gain a great advantage through the acceleration of the calculation of a greater number of shaders than ATI: all this still remains very theoretical. The calculation accuracy is 32 bits in floating point (s23e8) and a less accurate format is also available 16 bits floating point (s10e5). This additional accuracy provided by NVIDIA isn’t usually necessary, but it might be in very specific cases.

We have noticed that NVIDIA is using castrated GPU except with high end products that even include up to 512 MB of memory! Pixel rendering units are systematically inactivated. This might appear rather unusual for a relatively expensive professional graphic card. According to NVIIDA, in fact, it is only an additional security margin to enhance reliability. Note that the Quadro FX 3400 and 4400 uses a NV45, which is in fact based on a NV40 AGP and a PCI Express (not natively supported) bridge. You could buy three FireGL V7100 graphic cards for the price of one Quadro FX 4400. But once again, if the Quadro productivity improvement proves to be sufficient, it will still be interesting for some users whatever its price.

Test configuration:
Athlon 64 4000+
nForce 4 Ultra
Windows XP SP2
Drivers NVIDIA 71.24
Drivers ATI 8.08

We have added the GeForce 6600GT, GeForce 6800GT, Radeon X700Pro and Radeon X800XT score to have a performance comparison with standard graphic cards. Tests have all been made in 1600x1200.

Page 4
Synthetic tests


This first test shows the vertices´ transfer rate during basic geometrical calculations. Results are based almost exclusively on the number of vertex engines and the GPU frequency, and represent the raw calculation capacity. Logically with only 3 vertex engines and a frequency of 300 MHz, the Quadro FX540 is far behind. Performances are comparable whether if it is in T&L or in vertex shader.

This time three light sources are examined and a specular effect is calculated.
This increases the complexity of the operations to be processed on geometry.
This test shows NVIDIA´s supremacy in complex T&L process. Even the Quadro FX540, which has a lower amount of calculation unit, is able to provide performances close to the FireGL that contain a three times higher raw calculation capacity! NVIDIA has clearly optimised the vertex engines for T&L calculations that are very important for professional applications.

With calculation made in Vertex Shader 2.0, ATI comes out ahead. NVIDIA’s GPU still have issues with static branching, which should be faster than the dynamic equivalent. NVIDIA doesn’t seem to be in a rush to correct this problem. They have been aware of this problem for several months. One explanation could be that professional applications do not currently use very advanced vertex shaders, especially in DirectX and that video games are far from saturating geometrical units.

The third test is theoretical and shows the lighting calculation by pixel. The Quadro FX540 trails behind, but this is logical since it includes a reduced fillrate. Other graphic cards provide more or less equivalent performances. NVIDIA is slightly behind ATI for full precision and slightly ahead as regards partial precision, whether it is for simple or complex lighting.

When it is required to process a maximum of particles, ATI’s GPU are more efficient, thanks to a higher fillrate.

Page 5
Synthetic tests, next

Textures management
Fillrate in MPixels/s

By means of the Texbench utility, we have pushed all the graphic cards to deal with a great number of textures. We have chosen quadri-texturing of 1024x1024 textures. From 5 to 40 groups of 4 textures have been used to obtain 80 to 640 MB textures.
We can draw two conclusions from this figure. First: the performances that also include the GPU internal efficiency for accessing broad-ranging textures, and the mix of textures. Then: figures in red indicate that the rendering wasn´t smooth. Thanks to the 512 MB, the Quadro FX 4400 is the only graphic card to have been able to fluidly process 30 groups of textures (480 MB). The Quadro FX 3400 is the only one not to have been able to correctly process 320 MB of textures. The reason is probably due to the use of a bridge between the AGP supported by the GPU and the PCI Express slot. It is interesting to note that the GeForce 6800 GT doesn’t have this problem. This might be due to the 25% higher fillrate and the slightly higher bandwidth.

PCI Express

We have measured the transfer rate to send textures under Texbench for the 320 MB test. ATI graphic cards show an evident consistency and NVIDIA’s are clearly left behind. It is even more obvious for those equipped with a bridge. It is however still quite strange that NVIDIA’s native PCI Express solutions provide transfer rates so far removed from ATI´s graphic cards.

Serious Magic Texture Download doesn’t send textures but recovers some. The results this time are much closer between ATI and NVIDIA but the FireGL succeed in extricating themselves.

General Computing

Acoustics is a small demonstration unit using the GPU calculation capacity to calculate a room´s acoustics by means of an algorithm similar to the one used for ray tracing. The interaction with walls is of course taken into account. Most of this is chiefly calculated by means of the GPU and of the class 2.0 Pixel Shaders.

NVIDIA’s GPU clearly dominates ATI. Even the Quadro FX540 that included the lowest fill-rate of this roundup provided better than average results.

Page 6
3D Studio Max

3D Studio MAX

Specific drivers

3D Studio MAX is the referential 3D creation software. All professional graphic card manufacturers have specific drivers to replace the OpenGL or “standard” Direct3D. We have noticed that these drivers usually significantly improve performances. There are several additional options with this driver but not much of any importance. Of course these drivers only work with the Quadro and FireGL. For this test we used the specific driver for professional graphic cards, and the OpenGL driver for the Radeon and GeForce.

Line Anti-aliasing

Tests were carried out in both performance and quality mode. There were two main differences: texture filtering and line anti-aliasing. Here from left to right: we have a result without anti-aliasing, with anti-aliasing for ATI, and with anti-aliasing for NVIDIA. You will notice that the anti-aliasing quality differs from one company to the other. ATI applies more marked anti aliasing but simultaneously thickens lines.
There are pros and cons for both methods but we have a slight preference for NVIDIA’s. With this method, it becomes easier to distinguish the lines on more complex models whereas ATI’s method sometimes only creates a pile of indistinct lines.

Bugs : 1 each

ATI visibly likes big lines; sometimes a visualization section becomes the object of a strange anti-aliasing with something other than line thickening.

With NVIDIA’s specific drivers, the selected surface automatically turns white. It may possibly increase performances slightly but it also prevented us from distinguishing the surface correctly. This may well be due to a bug between the driver and 3D Studio MAX 7.


During the first part of these tests, ATI was dominant. We noticed an unusual result: general public graphic cards´ performances were similar whether the anti-aliasing was activated or not. The Quadro and FireGL were, however, significantly faster in performance mode. For the first test the GeForce turned out to be even 25% faster than the Quadro in quality mode. It is by no means easy to understand! Some optimizations are compromises, i.e. increasing performances in some situations but reducing them in others. This is probably the most logical answer.

For the second part of the tests that chiefly concern T&L, NVIDIA provided slightly better results. In fact all graphic cards provided a similar level of performances except for the GeForce 6600GT that got seriously left behind.

Page 7
3D Studio MAX, next, AutoCAD

3D Studio MAX,next

For this test, NVIDIA provided better performances with the FX4400 that simulated the graphic load of 3D Studio MAX. ATI and NVIDIA results are nonetheless contradictory: ATI is restricted by its geometrical calculation power and NVIDIA by the fillrate. All three ATI graphic cards provided very similar performances. The V5000 provide higher performances than the FX1400. The FX540 is mostly left behind, like all general public graphic cards.


We had initially planned to use the AutoCAD practice simulation tests for the AUGI Gauge test suite. Unfortunately, these practical cases were in the end entirely restricted by the CPU, which is representative of AutoCAD. We have however kept the AUGI Gauge synthetic test:

The time required to process the test is important, and it is interesting to note that the CPU is clearly restricting performances even if it allows the GPU some space to express its potential. ATI dominates this test but not by much. The Radeon are one step behind but the situation is different for the GeForce. The 6800 provides correct results but the 6600 provides the worse performances. The FX540 is slightly behind and on the same level as the Radeon

Page 8
CATIA, Cinema4D


This second test of the Viewperf suite clearly shows NVIDIA’s dominance: the FX3400 and FX4400 are 30% faster than their competitors. The FX1400 performances are once more restricted through the fillrate, but is nevertheless 50% faster than the general public graphic cards, all restricted by the GPU.

Cinebench 2003 is a benchmark based on Cinema 3D which provides two quite simple test scenes. Current GPUs shouldn’t have any problems with these scenes. :

Yes, it is a calculation time in seconds. An unidentified bug restricted the Quadro performances for the second scene, but not the GeForce despite the use of an identical driver. This bug made the Quadro 4x slower than they should normally have been. This problem isn’t present with the first scene and doesn’t concern the FX540 restricted by its fillrate. Performances are similar with all graphic cards.

Page 9
Ensight, Lightscape


With Ensight, ATI is a few percentage points ahead of NVIDIA. Once again, FX540 is clearly left behind and remains at the same level as general public graphic cards. It is interesting to note the clear advantage of Radeon over GeForce.


The situation is the other way around with the Lightscape. This time NVIDIA take the lead and is almost 20% faster. The FX540 provide great results almost similar to ATI’s graphic cards. General public graphic cards are a long way behind.

Page 10


NVIDIA once more takes the lead with the test of the Viewperf suite, and this time the performance gap was very impressive: +40%. The FX540 has nothing to be ashamed of compared to any of ATI’s graphic cards and the GeForce are in front of the Radeon. Only the FX1400 is a little bit behind. There might be two possible reasons for that: either the 128 Mb of memory restricts performances or else it doesn’t benefit from all optimisations included in other higher-end graphic cards, since it is clear that the driver created this performance gap.

We also carried out a few practical use tests with Maya, initially with the Spec dedicated benchmark group:

NVIDIA provided excellent performances with the Maya, even the GeForce were ahead of the FireGL. The Radeon were left behind, especially with the second test. You will notice the excellent performances of the FX1400, which can even beat higher end graphic cards in some of the tests.

Maya 6.5 introduced a high quality real time rendering mode using shaders written in Cg and that complied either toward extensions owned by NVIDIA for GeForce and Quadro, or else ARB extensions of all other GPU. We noted the impact on performances for several visualization modes:

From left to right: Wireframe, Wireframe + AA, Shaded, Shaded High Quality (the new mode).

If results are relatively close in the first two modes, even though NVIDIA had an advantage once the anti-aliasing was activated, the situation is completely different when the rendering becomes more complex. NVIDIA predominates clear over ATI and is twice as efficient. You will notice that there is no difference between professional and general public GPU. There is however a bug in the Radeon and the HQ scene (which doesn’t appear with the FireGL). The performance gap in this mode is really significant and it is better to have a NVIDIA graphic card to use it.

Page 11
Patchwork 3D

Patchwork3D developed by Lumiscaphe is a tool dedicated to high quality interactive presentation creation. This tool´s objective is to reach a level of quality far beyond any other real time application used in this test. Geometry and complex surfaces are used to reach this level of quality.

Unfortunately for ATI, this tool only works with NVIDIA’s GPU simply because developers didn’t trust ATI’s drivers. If we choose to include this application in our test, it is because it represents a part of the industry. Even if ATI’s driver quality has been improved, NVIDIA still is the reference and losing a bad reputation in the professional world takes a lot of time.

Nevertheless, Lumiscaphe told us that the reliability improvement of ATI’s drivers did finally convince them and that a new version compatible with ATI’s GPU was currently under construction.

While waiting for this new version, we have tested NVIDIA’s graphic cards with the current version. Lumiscaphe, whom we take this opportunity to thank, sent us three models for our tests, of increasing complexity. The first one uses 146155 polygons, the second 165207 polygons but this time with more complex surface and finally the last one gives a hard time to any GPU with its 449990 polygons.

Figures speak for themselves: the GPU really do have some difficulties. The poor GeForce 6600 GT collapsed and, even when the scene was working, it was unable to reach 1 FPS. Only the Quadro FX4400 was capable of reaching more that 10 FPS with the last model. The GeForce 6800 achieved similar performances to the Quadro.

Once in Wireframe render, performances improved clearly and the last scene performances were restricted by the CPU, except for the FX540. The GeForce 6600 was able to reach 1 FPS this time. The 6800GT isn’t able to catch up with the Quadro since they have drivers optimized by Wireframe render.

Page 12
Pro/ENGINEER, Unigraphics


Once more the FX 4400 is in the lead but this time the FireGL aren’t far behind and provide equivalent performances to the FX 3400. The FX 1400 and FX 540 are one step behind and general public graphic cards are far behind.

We also used this image to observe the render accuracy of complex scenes. With complex models a very high precision is required or else surfaces get mixed up and pixels flickers.

On the left NVIDIA, on the right ATI.

It is obvious that NVIDIA offers higher accuracy than ATI. The result is identical for all NVIDIA graphic cards and all ATI graphic cards. At first sight, with this little extract, it is possible to spot 10 anomalies for NVIDIA instead of more than 20 for ATI.


With the Unigraphics test for Viewperf, NVIDIA is once more in the lead and the FX 1400 is just behind the V7100. General public graphic cards are far behind especially for NVIDIA. The Radeon provided better results and even finished up in front of the FX 540.

Page 13


The situation is similar to the one of the Pro/ENGINNER test of Viewperf: the FX 4400 is in the lead followed by the FireGL and the FX 3400 is almost at the same level. The FX 540 is still behind but remains in front of general public graphic cards.

We also used the Spec practical tests for SolidWorks. We restricted results to 4 of them: the others are identical copies and don’t provide any additional information.

In more practical situations, NVIDIA clearly provides better performances than ATI, which is much slower in the Viewperf test, because even the FX 540 this time was in front of the FireGL. The 6800GT is also in front of ATI’s graphic cards and the 6600GT was placed within the first half of the tests.

Stereo mode

We installed a plug-in to activate the Stereo mode in SolidWorks. The image above is a monitor photograph, showing what we can see, if we look at a stereo image without the right system. To obtain more telling results, we emphasized this effect to the utmost, but that has no influence on performances.

NVIDIA is apparently more efficient than ATI in stereo mode with a performance cost of 45% instead of 50% for ATI. This performance gap isn’t however significant. However, once more NVIDIA is in front of ATI in terms of overall performances.

Page 14
Specific products, Conclusion

Personalised products
ATI and NVIDIA do not only develop professional products that anybody can find in shops. Both of them are able to provide personalized solution based on their GPU.
Obviously, these types of solutions are more expensive, since companies with very specific requirements generally require them.

An example, if you have looked at the pictures carefully, the FireGL V7100 PCB is able to include 512 MB even if no FireGL products with this amount of memory have been provided. The NV40/45 is able to support up to 2 GB of memory. It allows NIVIDA to meet some of the more specific requirements.

For quite some time now, ATI provides multi GPU solutions to Evans & Sutherland for flight simulators. NVIDIA announced, not so long ago, that Philips has contacted them for medical equipment. The graphic card professional market is simultaneously small and very far-reaching, this accounts for the price gap between general public, standard professional and specialized professional graphic cards.

Still on the subject of « options », NVIDIA has released a series of graphic cards with a Frame Lock and Genlock system to synchronize the image of several computers, in order to have a giant monitor. In other words, every professional market niche is interesting for ATI and NVIDIA and that makes sense since they generate nice profits and still manage to meet the customers’ requirements.

The same question keeps cropping up: is it better to buy an entry-level professional graphic card or a high-end general public graphic card for a computer used to play with, and to take the first steps in the 3D animation and modeling world. We feel that it is best to buy a GeForce 6800 GT, which has a broader range than an entry level professional graphic card such as the Quadro FX 540.

These graphic cards only come into their own in a professional environment and that is logical. The comfort gained thanks to improved performances, the options dedicated to graphic process application, the reliability improvement, and quality support are all more important factors than the quality/price ratio compared to general public products.

As an example: working on an important project without a smooth display of the application or with a lack of reactivity will soon reduce productivity, whereas a specially designed product, even if it is expensive and has a very low performance /price ratio compared to general public products, could solve these problems and provide a real bonus, whatever its cost.

Anyway, the initial target of this test was to decide between ATI and NVIDIA. There is no doubt that their professional range provides a bonus compared to general public products, but is a FireGL worth a Quadro, or vice versa? Our tests showed a strong domination of NVIDIA for performances and the NVIDIA OpenGL driver still benefits from industry recognition. There is no doubt that the best graphic card of this roundup is the Quadro FX 4400 and that the NVIDIA’s range generally performs better than ATI’s.

ATI has however a significant advantage: its price. Even if it is important not to keep in mind the usual price/performance ratio to evaluate these products. A FireGL V7100 is three times less expensive than a Quadro FX 4400! ATI seems to have decided to gain a few market shares by drastically reducing prices. All FireGL tested here include 6 vertex engines and this is the most important aspect for performances so it means that these graphic cards only cover a small part of performances. It also explains the small price range! The most interesting graphic card of the range is the FireGL V5000. It could be an even better product if it was equipped with 256 MB of memory and a more expensive power supply system on a level with its targeted market.

In the end it is clear that the Quadro FX 3400 and 4400 are only dedicated to a small number of users, who will choose them for specific reasons, including productivity and comfort gain. The Quadro FX 1400 and FireGL will meet the requirements for less specific performance searches, even if the choice for one or the other will have to be made according to the application used. For example, with Maya or a tool that only works with NVIDIA’s graphic card, the Quadro FX 1400 will be the obvious choice.

We will finish with a word about the FX 540 that gave us an overall good impression despite its situation compared to other graphic cards (which varies from the best to the worst). Thanks to NVIDIA’s extremely efficient drivers, it provides great performances in several situations but in other situation its entry level specifications aren’t enough to fill the gap. An entry level product is always a matter of compromise.

Copyright © 1997-2014 BeHardware. All rights reserved.