Professional Graphics Accelerators of Today: ATI vs. Nvidia

Today we are going to talk about those graphics cards that are intended not for gaming needs, but for serious work. We managed to get our hands on a few new solutions from Nvidia and ATI from the Quadro and FireGL series. Let’s find out what innovation shave they prepared for professionals working in 3D modeling applications and CAD software.

by Ilya Gavrichenkov
05/25/2006 | 04:45 PM

Some time ago we posted a review of the new professional ATI FireGL V7350 graphics card (for details see our article called ATI FireGL V7350 Graphics Card Review: Radeon X1800 XT for Professionals?). the launch of this solution based on the 90nm GPU was the first signal that the entire families of ATI’s and Nvidia’s professional graphics cards based on the latest and greatest GPUs will soon get a fresh new look. Of course, we couldn’t disregard this exciting bit of new from the professional graphics card market. So, today we would like to offer you a detailed coverage of features and performance of the latest solutions for professional workstations.

 

Note that even though the new-generation professional graphics accelerators use the same GPUs as the gaming cards, they still differ greatly from the gaming solutions. First of all it is a huge price difference that catches your eye: the professional solutions are a few times more expensive than the gaming cards. The price is justified by all other differences: the professional cards are targeted primarily at OpenGL applications for 3D modeling and CAD tasks for specialists. So, the special OpenGL driver specifically optimized for the corresponding software applications makes a significant part of the package. Moreover, all professional hardware and software from the graphics card makers is certified by the 3D application developers to ensure that they will work impeccably in these tasks. Therefore, we should regard professional graphics accelerators as a combo-package that includes not only the graphics card itself, but also a specially optimized OpenGL driver for this card.

Within our today’s test session we will take a closer look at the new ATI FireGL V series and Nvidia Quadro FX series solutions. These cards differ from their predecessors by much higher theoretical performance, which they owe to faster 90nm ATI R520 and Nvidia G71 GPUs. However, we should keep in mind that the workload created by 3D modeling and CAZD applications is completely different from the traditional gaming workload. Typical professional apps do not use shaders and other funky visualization technologies that have become extremely popular in the latest 3D computer games. Professional graphics cards need “pure” performance: texturing speed and triangle processing. Therefore, the use of GPUs initially developed for gaming needs may bring some surprises in professional environment. In other words, we can expect the performance of professional cards in OpenGL applications to strike most gaming fans as something quite illogical.

So, let’s meet our today’s testing participants.

Testing Participants

ATI FireGL V7350

We have already reviewed the top graphics card from ATI’s professional solutions family in our article called ATI FireGL V7350 Graphics Card Review: Radeon X1800 XT for Professionals?. Today this card is already selling for about $1,500, which is slightly less than the competitive solutions from Nvidia are offered for. However, FireGL V7350 is based not on the latest ATI GPU, but on the previous generation R520 chip. However, it seems to be offering the same texturing performance and triangles processing speed as the latest R580 GPU from ATI. In other words, both, R580 and R520, can lay 16 textures per clock cycle. The difference between them lies with the number of pixel pipelines. The new R580 has more pixel pipelines and hence works faster with pixel shaders. However, as we have already said, OpenGL applications hardly use any pixel shaders, so it doesn’t really matter that much. Despite this fact, FireGL V7350 is fully compatible with Shader Model 3.0 and even has 8 vertex processors.

The GPU of FireGL V7350 works at 594MHz, which is slightly lower than the GPU frequency of the Radeon X1800/X1900 XT gaming solutions based on R520/R580 chips.

The major advantage of FireGL V7350 over its predecessors and contemporary gaming graphics cards is the amount of onboard graphics memory it has. This professional graphics card appeared the first card out there with 1GB of GDDR3 SDRAM onboard. It works at 1,300MHz and with the 256bit bus delivers the bandwidth of 41.6GB/s.

The board has 2 DVI Outs, and thanks to Dual-Link technology each of them supports resolutions up to 3840x2400. Moreover, FireGL V7350 has the output connectors for 3D stereoscopic devices, such as glasses, for instance.

The card dissipated quite a bit of heat and is equipped with a dual-slot rotor cooling system. It requires additional power supply.

ATI FireGL V7200

Besides FireGL V7350, the professional graphics cards line-up from ATI also offers a few solutions will less memory onboard. Unfortunately, we didn’t get a chance to test the FireGL V7300 equipped with 512Mb of memory, however we got our hands on a FireGL V7200 with 256MB of memory.

The formal specifications of the FireGL V7200 are exactly the same as those of the FireGL V7350: the cards work at the same frequencies, feature the same number of pipelines and the width of their busses is the same also. Moreover, these two cards even look alike. You can see the difference only on the reverse side of the PCB: FireGL V7200 doesn’t have any memory chips there, unlike the top model. So, FireGL V7200 will be a solution for those professional users who do not need to store too many textures and who do not need a card with a lot of video memory.

As for the price point, the four times smaller amount of graphics memory has definitely had its positive effect on the price. Today FireGL V7200 sells in stores for about $800.

ATI FireGL V5200

FireGL V5200 professional graphics solution belongs to the “simple” graphics accelerators. While the FireGL V7XXX graphics cards are positioned for the high-performance workstations, FireGL V5200 is a mainstream professional OpenGL graphics accelerator. Its price indicates this positioning very clearly: $500 is a pretty medium price in this specific market segment.

However, if you have at least some idea about the situation in the gaming graphics cards market, you will be surprised to find out that FireGL V5200 is based not on the top ATI GPU, but on the mainstream solution. It is ATI RV530 manufactured with 90nm process and used for Radeon X1600 gaming graphics card family. RV530 GPU can lay only 4 textures per clock cycle, features 12 pixel and 5 vertex shader pipelines. Since this is a relatively new GPU, it is fully compatible with Shader Model 3.0.

FireGL V5200 graphics card is equipped with 256MB of GDDR3 SDRAM with 128-bit bus. The memory works at 1,380MHz. As for the GPU frequency, it works at 587MHz.

Just like the top professional graphics card models, the FireGL V5200 features two DVI Outs with Dual-Link technology support that allows connecting 9-megapixel monitors. This card doesn’t have the stereoscopic glasses connector.

Since RV530 is a much simpler GPU than R520, FireGL V5200 does perfectly well with a low-profile single-clot cooling solution, although it is an all-copper one. This graphics card doesn’t require additional power supply.

ATI FireGL V3400

FireGL V3400 professional graphics accelerator belongs to the entry level solution but costs around $300. From the gaming market prospective, it is a fully-fledged graphics card based on a 128-bit ATI RV530 GPU manufactured with 90nm process, just like the FireGL V5200. In other words, FireGL V3400 can boast 5 vertex shaders and 12 pixel pipelines that are fully compatible with Shader Model 3.0.

The major differences between FireGL V3400 and FireGL V5200 are the lower working frequencies and less onboard memory. Just like FireGL V5200, our hero features 4 texturing units, however its GPU works at 500MHz frequency.

As for the memory, FireGL V3400 features only 128MB of GDDR3 SDRAM working at 980MHz and supporting 128-bit bus.

In other words, FireGL V3400 graphics card has pretty similar hardware specs as the gaming Radeon X1600 PRO, while the more expensive FireGL V5200 is closer to Radeon X1600 XT.

Despite the relatively low price for a professional graphics solution, this card is also supplied with a special professional OpenGL driver that is optimized for major 3D modeling applications and cannot run on gaming cards.

As for the other features of FireGL V3400, it has 2 DVI Outs, one of which supports Dual-Link technology. So, you can not only use two displays (there can be analog displays as well as digital displays connected to it), but one of them can support resolutions up to 3840x2400.

I would like to say that FireGL V3400 looks very similar to FireGL V5200. Both these cards use the same PCB layout and the same cooling system. Moreover, you can only tell the difference between them by reading the memory chip marking, because they uses memory chips of different capacity.

Nvidia Quadro FX 5500

Nvidia responded to 1GB ATI FireGL V7350 launch with the release of their professional Quadro FX 5500 graphics accelerator. Besides the increase of onboard graphics memory to 1GB, Quadro FX 5500 has also become much faster than its predecessor, Quadro FX 4500, thanks to higher working frequencies.

The working frequencies of the new Quadro FX5500 as well as its architectural peculiarities make it resemble the GeForce 7900 GTX gaming solution with slower graphics memory. In other words, it uses a new 90nm G71 GPU that allowed Nvidia to raise not only the frequencies of their gaming solutions, but also the frequencies of their professional cards. Note that they managed to retain the same cooling system. Quadro FX 5500 features the same dual-clot heatpipe cooler as we saw on Quadro FX4500. As for the power consumption, the new card requires more power: the power consumption has been raised from 108W to 125W. Of course, the card requires additional power from the 12V lines of the system PSU.

The chip frequency of the Quadro FX 5500 graphics solution is 700MHz, while its memory (the card uses GDDR2 SDRAM) works at 1000MHz. so, only the GPU frequency has been raised compared with the Quadro FX 4500. The number of pipelines has also remained the same: the card has all 24 pixel and 8 vertex pipelines of the G71 architecture enabled. Of course, since the architecture of Quadro FX 5500 is similar to that of the top gaming products, Nvidia claim full compatibility with Shader Model 3.0. The memory bus is 256bit wide, so the peak bandwidth of the graphics memory makes 32GB/s.

Quadro FX 5500 features a pretty common set of connectors for the products of this type. There are two DVI and one stereo connector. Both DVI connectors certainly support Dual-Link technology that allows using displays with up to 3840x2400 resolutions.

Note that Quadro FX 5500 can boast some unique features that are absent by the competitor. Namely, Quadro FX 5500 cards can work in SLI mode and support Genlock and Framelock technologies (professional technologies for output synchronization to an external source and cluster visualization). These technologies are still in the planning stage for ATI FireGL solutions.

Nvidia Quadro FX 3500

Another new professional solution from Nvidia based on the latest G71 GPU is Quadro FX 3500. While the top-of-the-line Quadro FX 5500 costs around $2,000, Quadro FX 3500 sells for as low as $1,000, although it uses the same GPU. The prices of these two graphics solutions are so far apart, because the differences between the Quadro FX 5500 and Quadro FX 3500 are much more significant than just clock speeds and smaller amount of onboard graphics memory.

First of all, when we speak of G71 GPU in Quadro FX 3500 we should keep in mind that Quadro FX 3500 is based on a cut-down modification of this chip. Some pipelines of this GPU in Quadro FX 3500 are disabled, so its only has 20 pixel and 7 vertex units. At the same time, the GPU clock speed of this card has been reduced to 470MHz. So, Quadro FX 3500 is evidently much slower than the high-end professional Quadro FX 5500 from the theoretical performance point of view.

Quadro FX 3500 is equipped with 256MB of GDDR3 SDRAM connected to the GPU via a 256bit bus. The memory on this graphics card runs at 1320MHz, which makes this solution a memory bandwidth leader of the entire Quadro FX family.

The maximum power consumption of Quadro FX 3500 based on a cut-down G71 GPU is 80W. This allows using a single-clot copper cooling solution. Nevertheless, this card does require additional power supply for proper operation.

Just like its elder brother, Quadro FX 3500 features two DVI Outs with Dual-Link technology support on both of them and a stereo out. Moreover, the card supports SLI technology. As for Genlock and Framelock technologies, they are not implemented on this solution.

I have to point out that there are no cards with 512MB of graphics memory among the new professional solutions from Nvidia. However, the launch of the new Quadro FX 5500 and Quadro FX 3500 cards doesn’t at all mean that they will out the predecessor, a great Quadro FX 4500, from the market. This popular professional solution is based on G70 GPU and features exactly 512MB of GDDR3 memory. It will remain in the line-up of Nvidia’s professional graphics accelerator family.

Nvidia Quadro FX 1500

Another professional graphics card based on G71 GPU, used for GeForce 7900 gaming graphics accelerators, is Quadro FX 1500 with the recommended price of $700. G71 chips used in this solution are even more modified than the ones in the previous case. These GPUs feature only 16 pixel and 6 vertex pipelines. The GPU frequency of Quadro FX 1500 is only 375MHz.

However, this artificial slowing down made Quadro FX 1500 less power hungry and it doesn’t require any additional power supply. This is a great distinguishing feature of this solution, because other than that it is exactly the same as the Quadro FX 3500 we have just discussed. The PCB is equipped with the same cooling solution and the memory chips are also the same.

The use of identical memory chips means only one thing: Quadro FX 1500 has 256MB of GDDR3 SDRAM and supports 256bit memory bus. The memory of this card works at 1250MHz.

Quadro FX 1500 has the same two DVI Outs as Quadro FX 3500, and supports the same technologies. There is only one difference though: it features an HD Out instead of the stereo output, therefore you can connect an HDTV received to it.

Nvidia Quadro FX 560

Quadro FX 560 is the youngest solution in the professional graphics card family from Nvidia. It sells for about $270. It is based on G73 GPU that is also used for GeForce 7600 and 7300 gaming cards.

The GPU modification used in Quadro FX 560 features 12 pixel and 5 vertex pipelines. The chip frequency equals 350MHz. It is equipped with 128MB of graphics memory and the memory bus is 128bit wide. The graphics memory works at 1200MHz frequency.

Quadro FX 560 features its own specific PCB layout. The GPU is covered with a small aluminum heatsink. Of course, no additional power supply is necessary for this solution. The card features two DVI Outs and an HDTV Out. The Dual-Link technology is supported by only one DVI Output.

Note that even though the manufacturer doesn’t officially claim that Quadro FX 560 supports SLI technology, the corresponding connector is there, so it gives us some hope that we will be able to enable this technology on Quadro FX 560 cards one day.

Testbed and Methods

We tested our professional graphics cards on a workstation built around a dual-core AMD Athlon 64 FX-60 processors and featuring 2GB of RAM. Here is a detailed list of hardware we used to build our workstations:

These are the drivers that we used for our professional graphics cards test session:

We tested the cards in professional applications in 1280x1024 resolution with Vsync disabled unless stated otherwise. Before measuring the performance in all applications we selected the corresponding profile in the OpenGL driver.

Comparative Specifications of Testing Participants

Before we pass on to discussing the benchmark results let’s take a closer look at the comparative chart for the specifications of the testing participants. This will help us make the right conclusions from the results we are going to see below:

Note that it is not an easy task to compare formal specifications. The thing is that ATI and Nvidia have absolutely different architectures that is why even if there are more functional units and the clock speed is higher it doesn’t necessarily indicate higher performance in real applications. Moreover, drives have great influence on performance of professional solutions. However, we still have to stress that if we compare the peak theoretical performance, then the top Nvidia solutions will be about 1.5 times faster than ATI solutions.

Performance

SPECviewperf 8.1

First of all we decided to test the performance of our top professional testing participants in SPECviewperf 8.1 benchmark, which serves as a sort of reference when it comes to graphics cards performance in 3D modeling tasks. This benchmark shows the performance during rotation of typical models created in various professional applications. Note that this benchmark doesn’t allow us to evaluate the performance of our testing participants in each application separately. Since the developers do not optimize their drivers specifically for SPECviewperf 8.1, this benchmark can be regarded as a great tool for “raw” performance analysis of the graphics card and the driver in OpenGL.

The results obtained in SPECviewperf 8.1 are given on the diagrams below:

In SPECviewperf 8.1 test professional solutions from Nvidia Quadro FX family boast higher results. This is not surprising at all, because these graphics cards boast higher “raw” theoretical performance thanks to more pipelines onboard.

However, we wouldn’t take the SPECviewperf 8.1 results too close to heart. They may be not representative of the performance in real professional applications. Therefore, let’s measure the performance of our testing participants in most typical 3D modeling tasks. We used SPECapc test scripts for our tests.

3ds max 7

According to the total score from SPECapc test, Nvidia Quadro FX graphics cards are much faster in 3ds max than their ATI competitors. I would like to stress that we are observing a pretty funny situation with Quadro FX 3500 being faster than the more expensive Quadro FX 5500. It means that 3ds max is sensitive not only to the GPU speed, but also to the graphics memory frequency, and Quadro FX 3500 boasts the highest memory bandwidth of all the Nvidia professional solutions. I would also like to point out the low result of Quadro FX 4500. Of course, this is a previous generation OpenGL card, which - alongside with the low working frequencies - explains its low performance.

The charts below show more details results for the professional graphics cards performance in 3ds max 7:

ATI FireGL graphics accelerators do not always lose to Nvidia Quadro. As we can see, when we work with reframe models ATI FireGL solutions run at a pretty competitive speed. However, since they do not run fast enough under any other types of workload, we cannot give these cards a positive verdict.

Maya 6.5

The situation in Maya is completely different from what we have seen in 3ds max 7. ATI professional solutions are indisputable leaders here leaving Nvidia Quadro competitors far behind. Even the slowest FireGL V3400 outperforms the top Quadro FX model. We suspect that there are some problems with Nvidia drivers, which caused these low results in Maya tests. As we can see from the diagrams, the performance of entire Nvidia Quadro family appears limited by some value, which origin we cannot explain at this time.

Besides the total scores in Maya we would also like to offer you the results of the corresponding SPECapc test during work with corresponding test models.

These results show once again that the performance of Quadro FX graphics cards stumbles over some issues. And it is definitely not the CPU that is to blame, because FireGL accelerators easily leave their competitors behind, and sometimes this advantage is measured in times.

By the way, FireGL V7350 is almost always slower than the less expensive FireGL V7200. These two solutions differ only by the amount of onboard graphics memory, which doesn’t seem to always help. Although, the lag is not that dramatic.

SolidWorks 2005

The performance in SolidWorks 2005 doesn’t depend as much on the power of the graphics accelerator as the results of other 3D modeling applications. But nevertheless, we can still conclude that Nvidia Quadro FX cards are ahead. These graphics cards show the best performance results.

And it is evident not only from the average benchmark scores but also from the more detailed performance reports for work with particular details of the app.

Note that the results obtained for the test working with the engine model are slightly off the route. This gives us to understand that Nvidia Quadro FX cards are not always a preferred choice for SolidWorks 2005. I would also like to point out that the graphics memory speed has quite a bit of influence on the performance in this application. Therefore, Quadro FX 3500 graphics card may sometimes be faster than Quadro FX 5500.

AutoCAD 2006

Once again Nvidia Quadro FX graphics cards are ahead of ATI FireGL in this professional OpenGL application. The total scores you see on the diagram above are explained by the detailed reports:

There is only one surprise here: 2D performance. We would evidently expect it to be pretty similar by all testing participants. However, the low result of FireGL V7350 really stands out here. It is probably 1GB of graphics memory that for some reason affects the 2D performance of the card. At the same time, I would like to stress that the previous generation Quadro FX 4500 is also quite fast in AutoCAD 2006. Since it yields to the next generation solutions in most other applications, here its results look more than decent.

PRO/Engineer Wildfire 2.0

Although the results of PRO/Engineer Wildfire 2.0 are not really dependent on the power of the graphics accelerator, Nvidia professional solutions are again performing better.

CINEBENCH 2003

Besides the SPECapc family of applications we have also resorted to a special CINEBENCH 2003 test. It shows the systems’ performance in a popular Cinema 4D application, which is especially beloved by Mac users.

Nvidia cards are evidently faster here. Moreover, we even got the impression that our CPU prevented the Quadro family from showing its real power.

3DMark06

Although professional graphics accelerators are hardly ever used for games, we decided to still run the FutureMark 3DMark06 test. The thing is that unlike 3D modeling applications, this benchmark allows to better estimate the computational power of the GPU used for shader processing.

This is where we see clearly that the gaming workload is absolutely different from the workload created by professional OpenGL applications. The performance of our testing participants in 3DMark06 doesn’t resemble anything we discussed today. As we have already mentioned, this test doesn’t care about the triangles processing and texturing speed: all it needs is high shader performance, and we know that shaders aren’t that widely spread in professional tasks.

Conclusion

We have just taken a real close look at the new generation of professional graphics accelerators from ATI and Nvidia based on the new GPUs manufactured with 90nm process. I would like to stress right away that ATI and Nvidia cards acquired a significant performance boost thanks to the new R520, RV530, G71 and G73 GPUs used for them. Although the new GPUs are primarily remarkable for their faster shader processing speed, more advanced production process allows the developers to increase the number of pixel pipelines and raise the working frequencies. As a result, we got an impressive performance growth in 3D modeling and CAD applications.

Moreover, the top ATI FireGL V7350 and Nvidia Quadro FX 5500 graphics cards got 1GB of graphics memory this should definitely be demanded by some professional users who work with huge amounts of texturing data.

I would like to draw your particular attention to mainstream professional solutions that proved very interesting. We were especially happy with the Nvidia Quadro FX 3500. Thanks to its fast memory and G71 GPU this solutions demonstrated superior performance and sometimes proved even faster than its high-end brother.

Its rival, ATI FireGL V7200, also performed quite well in our test session. It demonstrated the same performance as its more expensive fellow – FireGL V7350. So, if you do not need 1GB of graphics memory, and are happy with ATI’s professional graphics solutions, then FireGL V7200 will be a great choice for you.

Summing up, I wouldn’t give any specific recommendations about purchasing this or that professional product. These solutions are quite expensive and you should address this matter responsibly. Especially, since professional users are very rarely working in several different applications at the same time. Therefore, your choice of a workstation graphics card will be strictly determined by the type of applications you will be running on your platform. Of course, Nvidia Quadro FX solutions are faster in most cases than ATI FireGL, however there are tasks where FireGL will be irreplaceable. For example, in Maya. So, we encourage those of you who have read this article to the end to make your own conclusions basing on the results and analysis we have shared with you today.