by Ilya Gavrichenkov
09/26/2007 | 09:24 AM
Most articles you can find in the Video section of our website are one way or another connected with the gaming graphics solutions. Reviews of these products are of interest to majority of our readers in the first place. However, the graphics card market is not limited only by the new gaming products, of course.
There is another class of graphics accelerators, evolving in parallel to the gaming world, that are targeted for specialists and not hardcore gamers. These are the so-called professional solutions designed to be using in 3D modeling and automatic computer-aided design (CAD) platforms. We do not write very often about solutions like that: most professional do not base their decisions on the performance tests, they rather look at other criteria. For example, they trust the software developers’ recommendations, claimed supported features and even their own personal experience more than performance tests.
Nevertheless, we do our best not to forget about solutions like that and are proud to offer you detailed reviews of these products on a regular basis. Luckily, the new solutions do not come out very often in this field and the release of new models is often tied up to the new GPU architecture announcement.
This time our professional graphics cards discussion was provoked by the fact that relatively new Nvidia Quadro FX 5600 and Nvidia Quadro FX 4600 professional graphics accelerators, launched in March of 2007, finally became available in retail. These solutions, just like GeForce 8800 graphics card family, are based on chips with G80 graphics architecture with 128 streaming processors.
So, our today’s article will be devoted to the features and performance of these two professional products.
It is true, it may seem at first glance that professional graphics solutions are not so widely spread. Their high price of a few thousand dollars per card and absence of traditional advantages typical of gaming solutions may scare away even experienced users. Nevertheless, the professional graphics card market not only continues its successful existence, but is gradually growing over the past few years.
Professional graphics accelerators are usually used in high-performance graphics workstations in very diverse fields of industry. I would like to note that the general notion of “professional graphics solution” doesn’t provide a definite product description. In fact, the main attribute of professionalism may be the specific positioning of the solution for special applications used by high-end professionals. Therefore, we can easily assign the “professional” tag not only to the 3D graphics accelerators reviewed today that serve for work in digital content creation, 3D modeling or CAD applications, but also to the 2D accelerators that may be used in television, non-linear video editing and in stock exchange systems. However, 2D graphics cards are not so interesting from the technological standpoint so we usually do not review them.
The users believe the main difference between 3D gaming and professional applications to be the API they employ. Most gaming applications work via DirectX, while professional 3D modeling and CAD applications use mostly OpenGL. At the same time, there are also OpenGL games, such as Quake 4, for example. However, the FPS rate in this game doesn’t in any way correlate with the professional applications performance, which indicates dramatic differences between gaming and professional applications. Top gaming graphics cards that can easily deliver excellent gaming performance in Quake 4 and other OpenGL games with maximum image quality, may turn out absolutely helpless in 3ds max, Maya, Lightwave or AutoCAD. And the problem is not in the OpenGL API these applications employ, but in the type of workload created by professional applications that is completely different from the one created by games.
No matter how crazy it sounds, but all scenes in contemporary computer games are relatively simple. Judging by the professional applications standards they consist of comparatively few polygons that, moreover, do not change their position in relation to one another that much. And the impressive looks of the gaming applications comes primarily from colorful textures and different effects, including shader effects, of course.
Professional applications load graphics accelerators in an absolutely different way. The scenes employed by 3D engineers and designers usually use by one, two or even three orders of magnitude greater number of polygons. However, professional applications hardly use any effects or even texturing. Professionals fell more like working with wireframe models or Gouraud shading leaving the fully-fledged image to the final rendering stage that is performed by the system CPU and doesn’t at all depend on the graphics card capacity. Moreover, professional applications require high geometric performance of the graphics accelerator, which is essential for deformation of objects that are being frequently used by the professionals.
So, the key parameter for professional graphics accelerators is the geometric GPU performance, while for the gaming graphics cards texturing speed and shader unit performance are more important. Besides, professional accelerators may often have hardware anti-aliasing support and a number of other similar functions aimed at improving the display of 3D models that aren’t required by gaming solutions and hence are absent there.
Another additional peculiarity of professional graphics solutions is the special OpenGL drivers optimized for corresponding applications. Professional graphics adapters are designed for specific tasks, which allow applying slightly different approach to driver development. In this case not only the performance during polygonal models processing, but also their impeccable display matter the most.
In other words, designer, 3D artist or animator needs to see the correct picture on the monitor, displayed without any omissions or simplifications that often occur in games for the sake of acceptable FPS rate. It means that professional graphics card drivers will never sacrifice quality for performance – this is their principal difference from their gaming 3D accelerator drivers. Besides, professional hardware and software solutions from graphics card manufacturers need to be certified by the 3D application developers in order to ensure that they will work flawlessly in these applications at all times.
This way, professional graphics accelerators may be regarded as a unified hardware-software complex that includes not only the graphics card itself but also an OpenGL driver specially optimized for particular applications as guaranteed by the developer of professional 3D software suite.
Besides everything we have already said, contemporary professional graphics cards acquired a number of features that are absence by the gaming graphics solutions. Most professional cards, especially the ones from the high-end category, are equipped with a slightly different set of output ports and connectors. First of all, I would like to say that DVI ports on professional graphics cards often support Dual-Link technology. Thanks to this technology, they can support 9-megapixel monitors with gigantic resolution of 3840x2400. Secondly, professional graphics cards often come equipped with an additional stereo out for stereoscopic glasses that have never become widely spread in gaming environment.
It is also important that the latest professional graphics card generation support GenLock and FrameLock technologies. The first one serves to synchronize the signal generated by the graphics card with the external source, which is used a lot in video transmitting applications. The second one implements cluster visualization principles and allows independently synchronizing the channels from numerous workstations and creating a large virtual display managed by the numerous systems cluster for the sake of higher performance.
Now that we have introduced to you the application and general peculiarities of professional graphics it is time to take a closer look at the actual testing participants of our today’s review: Nvidia Quadro FX 5600 and Nvidia Quadro FX 4600 graphics cards.
Although the digital index of the Quadro FX 4600 model is not too high, this graphics accelerator, just like Quadro FX 5600, belongs to the high-end solutions. No wonder, the formal specifications of Quadro FX 4600 card are very close to those of GeForce 8800 GTS. In fact, it even uses a similarly designed PCB 229mm long.
Quadro FX 4600 is based on G80 chip manufactured with 90nm process. It works at 500/1200MHz (these are frequencies of the geometric and shader domains respectively). So, the graphics card supports DirectX 10 and Shader Model 4.0. Just like the gaming counterpart, Quadro FX 4600 has fewer, 96, streaming processors. So, the theoretical triangles processing speed of this graphics card is 250 million triangles per second and the fillrate reaches 12 billion texels per second.
However, the analogy between Quadro FX 4600 and GeForce 8800 GTS is not complete. Unlike its gaming relative, Quadro FX 4600 is equipped with 768MB GDDR3 SDRAM that allows employing fully-fledged 384-bit memory bus on this card. In this case the memory works at 1400MHz frequency providing 65.8GB/s bandwidth.
The maximum power consumption of the Quadro FX 4600 is 134W, so the card can do with only one 6-pin power supply connector.
Quadro FX 4600 borrowed the design of its cooling system from the gaming graphics cards. It uses a standard dual-slot cooler including a massive aluminum base, a heatsink array of thin ribs, a heatpipe, a centrifugal fan and a transparent plastic casing. This cooling solution proved highly efficient and pretty low-noise even in gaming cards that is why we were not at all surprised to see Nvidia employ it for their professional graphics card series as well.
Quadro FX 4600 comes equipped with a pretty standard set of connectors for the solution of this class: two DVI connectors and one stereo connector. Of course, both DVI ports support Dual-Link technology that allows connecting two monitors with up to 2560x1600@60Hz resolution.
It is important to point out that Quadro FX 4600 may also work in SLI configurations that are supported in the hardware as well as in the drivers. Also this card theoretically supports GenLock and FrameLock (professional image synchronization with the external source and cluster visualization technologies), however the Nvidia Quadro G-Sync daughter card required for their proper functioning is not available yet, even though there is the corresponding connector at the upper edge of the PCB.
Nvidia Quadro FX 4600 sample we received for our tests was made by PNY Company. However, it doesn’t actually matter, as all professional graphics cards on Nvidia chips are anyway manufactured on one and the same facility.
So, PNY is responsible only for the packaging and accessories bundle that includes two DVI→D-Sub adapters and a power adapter for the 6-pin power connectors on the PCB. The recommended price of this solution is set at $1,995.
The today’s fastest professional graphics card is Quadro FX 5600. It not only works at higher clock speeds, but also boasts maximum amount of onboard graphics memory available today – 1536MB. There are no gaming graphics accelerators that could boast that much onboard graphics memory today. However, this situation is quite logical. Graphics cards with 1GB of memory at first appeared in the professional segment.
As you may have already realized, Quadro FX 5600 has no analogues in the gaming segment. The card features unique PCB and unique cooler design that may make an unforgettable impression due to their size. The thing is that Quadro FX 5600 is a full-size graphics card, it is 312mm long. Besides, it is also equipped with a steel railing on the farther side to ensure its secure retention inside the case and is enforced with an armed steel panel attached along its upper edge.
As for the technology behind Quadro FX 5600, it is based on the same 90nm G80 chip overclocked to 600/1350MHz frequencies (the geometric and shader domain speeds, respectively). So, is we still try to draw some analogies with the gaming solutions, we will be able to say that Quadro FX 5600 yields a little bit to GeForce 8800 Ultra from the processor speed standpoint. Nevertheless, the theoretical performance of Quadro FX 5600 looks very impressive. Thanks to the full set of 128 streaming processors the card can draw 300 million triangles per second and texture 19.2 billion texels per second. Of course, it is fully compatible with DirectX 10 and meets Shader Model 4 requirements.
1.5GB of GDDR3 SDRAM installed on this board is clocked at 1600MHz and use standard for G80 384-bit bus. So, the peak bandwidth of the video memory subsystem equals 76.8GB/s.
Just like the top gaming graphics cards built around G80 GPU, Quadro FX 5600 consumes quite a lot of power – 171W. That is why it is equipped with two 6-pin power connectors.
To cool down the graphics processor and memory Quadro FX 5600 uses a monstrous cooler that is even larger than the one used on GeForce 8800 Ultra cards. However, the cooler design has no specific peculiarities: it simply looks like a stretched out Quadro FX 4600 cooler with longer ribs in the heatsink array.
Quadro FX 5600 comes with the same set of connectors as Quadro FX 4600: two Dual-Link DVI and a port for stereoscopic glasses.
Just like its younger brother, Quadro FX 5600 supports SLI technology and has a corresponding connector onboard. There is also a special connector for Nvidia Quadro G-Sync card implementing GenLock and FrameLock functionality.
The card was provided by PNY Company and comes with a pretty standard bundle of accessories:
The box with the graphics card also contains a user’s manual, a CD-disk with drivers, two DVI→D-Sub adapters and two power adapters. The recommended street price for Quadro FX 5600 is set at $2,999.
Nvidia offers a special Forceware driver package for their Quadro graphics cards series. It is very similar to the driver suite for gaming solutions. The driver suite comes with a traditional control panel that offers professional profiles instead of gaming ones.
Before we pass one to the actual benchmark results in professional graphics applications we need to sum up their technical specifications in a single table for your convenience. It will help us interpret the practical results correctly.
Take a look:
We decided to compare the new professional Nvidia graphics cards, Quadro FX 4600 and Quadro FX 5600, against the previous generation top professional graphics solutions – Quadro FX 5500 (you can check out our detailed test results of this solution in the article called Professional Graphics Accelerators of Today: ATI vs. Nvidia). Moreover, we were also going to compare the performance of the new Nvidia solutions against that of the newest competitors from AMD/ATI. However, AMD Company failed to provide us with the recently announced ATI FireGL V8650 and ATI FireGL V7600 graphics cards. Therefore, today we will only compare the new Nvidia solutions against the previous generation ATI FireGL V7350 and the review of the latest FireGL solutions will hopefully follow shortly.
We tested professional graphics cards on a workstation built around quad-core Intel Core 2 Extreme Q6850 processor working at 3.0GHz frequency and equipped with 2GB of RAM. Our workstation was configured as follows:
The tests were run with the mainboard BIOS Setup adjusted for maximum performance.
This is the list of the drivers we used during our professional graphics card test session:
The graphics cards were tested in professional applications in 1680x1050 resolution with Vsync disabled, unless stated otherwise.
First of all we decided to check the results in the main professional graphics card test - SPECviewperf benchmark, which newest 10th version has been released recently. Although SPECViewperf is a synthetic test, it serves as a certain reference when it comes to graphics cards performance in 3D CAD and modeling applications. This benchmark transmits to the graphics card driver a pre-formed succession of OpenGL commands typical of different professional applications. The built-in scripts emulate the user’s work in viewports of the following applications (the names of the corresponding benchmarks are given in brackets): 3ds max (3dsmax-04), CATIA (catia-02), EnSight (ensight-03), Maya (maya-02), Pro/ENGINEER (proe-04), SolidWorks (sw-02), UGS Teamcenter Visualization Mockup (tcvis-01) and UGS NX (ugnx-01).
The obtained results are pretty evident: the new generation graphics cards are faster than the older offerings from AMD and Nvidia. Moreover, Nvidia Quadro FX 5600 with higher clock speeds outperforms Nvidia Quadro FX 4600.
The new version of SPECViewperf test also allows testing full-screen antialiasing. Its final report reveals the types of antialiasing that can be used in different applications without losing much of the performance.
NVIDIA Quadro FX 4600:
NVIDIA Quadro FX 5600:
another report from SPECViewperf 10 shows how scalable the video subsystem performance is during work with multi-threaded graphics content.
NVIDIA Quadro FX 4600:
NVIDIA Quadro FX 5600:
Besides SPECViewperf 10 we have also tested our professional graphics accelerators in a number of real 3D modeling suites with the SPECapc test scripts.
The peculiarity of 3d max 9 tests is that it uses not OpenGL but DirectX. They recommend this particular API for the latest versions of this popular 3D modeling suite.
New Nvidia graphics cards demonstrate the best results here, too. Note that Nvidia Quadro FX 5600 shows the best shader performance, which is quite logical, while the best general graphics score belongs to Nvidia Quadro FX 4600.
More detailed results of the 3ds max 9 performance test reveal the advantages of the new Nvidia professional solutions:
As we see, high frequencies and more graphics memory of Nvidia Quadro FX 5600 do not always help this card win the first prize in the tests. It would be of interest primarily to those professionals who deal with large textures a lot. In many other cases Nvidia Quadro FX 4600 may be a more attractive solution only from the performance-per-dollar prospective.
Although new Nvidia accelerators work much faster in Maya 8.5 they still cannot outperform the previous generation ATI card in the total score chart. However, Nvidia Quadro FX 4600 and FX 5600 didn’t lose in every single test: in some cases their performance proved pretty impressive.
Note that again Nvidia Quadro FX 4600 sometimes looks better than the Quadro FX 5600 solution.
The results indicate that the performance of top professional graphics solutions from Nvidia, Quadro FX 4600 and Quadro FX 5600, hit against the limitations of the system CPU. That is why they demonstrated similar results, although the highest anyway.
Although professional accelerators are hardly ever used for gaming, we decided to include the results of the gaming FutureMark 3DMark06 testing suite. The thing is that unlike modeling applications, this benchmark can best of all estimate the computational potential of the GPU involved in active shader processing.
Here we finally reveal the fact that gaming workload is dramatically different from the workload in professional OpenGL applications. The performance picture in 3DMark06 is completely different from what we have seen before, because this test is sensitive not that much to triangles drawing or texturing speed, but mostly to processing of shaders that are barely used in professional graphics at all.
Here we are going to discuss the measured power consumption level of complete systems (without the monitor) equipped with different professional graphics accelerators. The measurements were taken in 3dsmax-04 test from the SPECViewperf 10 suite.
The monstrous Nvidia Quadro FX 5600 graphics card proved to be the most power-hungry solution of all our testing participants. Nvidia Quadro FX 4600 is much more economical, however the previous generation cards required even less power than that.
As we have expected, the transition of Nvidia’s professional graphics card family to the new G80 chips raised their performance to a new level. Nvidia Quadro FX 4600 and Quadro FX 5600 graphics cards perform much faster in professional applications than the previous generation accelerators.
As for the direct comparison between the Quadro FX 4600 and Quadro FX 5600, the first card from this pair made a very good impression. Its performance is more than sufficient for successful work in all major professional applications and in some cases it is even faster than the top solution. Moreover, this card is more economical and less expensive than Nvidia Quadro FX 5600. The major advantage of Nvidia Quadro FX 5600 is actually not its unprecedented performance. This graphics card will be of interest to specialists first of all thanks to 1.5GB of onboard memory. There are no other solutions that could boast that much memory in Nvidia’s as well as competitor’s product range. Therefore, this feature makes it irreplaceable for those cases when you need to work fast with large textures.
Unfortunately, we cannot yet compare the Quadro family with the new professional graphics cards from AMD, but we hope that their promising professional products will arrive in our lab shortly. This is when we will make the final conclusions about the leadership in the professional graphics card market. So, stay tuned!