by Ilya Gavrichenkov
07/27/2010 | 05:33 AM
Nvidia has been steadily improving its standing on the market of gaming GPUs. Having struggled to put early Fermi-based graphics cards into mass production, the company found that, despite their undoubtedly high performance, they met but a lukewarm reception among users due to their noisiness, high heat dissipation and costliness. Nvidia kept on pushing its Fermi family on the gaming market, though, and today offers mainstream graphics cards with a simplified Fermi chip which is free from most of the shortcomings of the full-featured version. Nvidia is also preparing even cheaper DirectX 11 compatibles for the back-to-school season. Thus, it will have become quite competitive to AMD across the entire DirectX 11 product line-up by the end of this year but it must be noted that the troubles with the introduction of the new generation of gaming GPUs have cost Nvidia over 10 percent of its market share.
On the other hand, there is a market segment where Nvidia has been on the rise. We mean professional graphics cards that serve in 3D modeling and computer-aided design workstations. Why? Well, professional designers, modelers and engineers need high-performance graphics hardware, just like gamers, and sometimes they need even more performance than computer enthusiasts. However, a professional’s notions about graphics card performance are somewhat different from that of ordinary users.
Professional graphics cards are not required to create spectacular and picturesque visuals with eye-catching shader effects, explosions, water rendering, and dynamic tessellation. They usually don’t even need to quickly draw a scene with nice-looking textures. Instead, graphics cards for CAD and CAM workstations must deliver high performance from a different point of view. They must be able to efficiently process huge arrays of geometrical data using some basic operations. A designer, engineer or architect needs to load into his graphics card a model with a huge number of graphical objects but in most cases the model is going to be processed in wireframe mode or using simple texturing and lighting. Rotation is going to be the most popular operation to be performed over such a model.
Thus, it looks like all the new-fangled technologies of photorealistic visualization touted by the GPU developers do not matter for professionals. Therefore, the lack of DirectX 11 compatibles in Nvidia’s professional product line-up has not prevented the company from increasing its market share. Again, engineers and designers need different things than gamers. They are willing to pay a lot of money for high-quality technical support, for polished-off and problem-free drivers, and for guaranteed stability of graphics cards in their CAD/CAM systems. Nvidia’s professional products have been very good from these aspects in the last few years. Besides developing an independent driver series for such solutions, the company optimizes them for popular CAD suites, offering specialized software.
However, we don’t mean that this market segment develops at a slower rate than the market of gaming graphics cards. Using the same GPUs, professional cards change their generations at about the same rate as gaming ones, but the introduction of a new generation is somewhat delayed due to the need to optimize the drivers and certify the products with CAD and CAM suite developers.
Fermi-based professional graphics cards were sure to come out after their gaming counterparts and now the time has come. Nvidia is announcing a series of new Quadro cards based on the well-known GF100 chip. The new series includes three models belonging with different price categories: the Quadro 4000 comes to replace the Quadro FX 3800; the Quadro 5000 is a replacement for the Quadro FX 4800; and the top-end Quadro 6000 is the new-generation substitute for the Quadro FX 5800.
Nvidia has been kind to offer us a sample of the most exciting product of the new series. So we will tell you all about the new Quadro 5000 in this review.
If you have been keeping in touch with the market of professional graphics cards, you should already know that products for high-performance workstations are hardly any different from their gaming counterparts. AMD and Nvidia have long developed unified GPU architectures and use them for any applications. By the way, this approach helped them oust the companies that used to develop and promote specialized GPUs for professional applications. Thus, each Quadro series card has a certain gaming cousin, and the new series is no exception.
On the other hand, the Quadro 5000 doesn’t have an exact gaming copy. Although based on the GF100 chip, it has certain unique features in its hardware specs. The execution domain of the GF100 GPU is known to include 16 multiprocessors, each with 32 stream execution cores, but no gaming card makes use of all these resources.
The same is true for the Quadro series. Even the senior Quadro 6000 has only 448 active shader processors out of the total 512. On the other hand, Nvidia enhanced the card’s memory subsystem which is especially important for professional products that have to process huge models. The Quadro FX 4800 used to come with 1.5 gigabytes of memory whereas the Quadro 5000 has as much as 2.5 gigabytes. There is no gaming graphics card today to have that much of graphics memory, by the way. Clocked at 3.0 GHz and accessed via a 320-bit bus, this memory subsystem provides a peak bandwidth of 120 GBps. As a result, the geometry-processing performance of the Quadro 5000 is lower than that of the GeForce GTX 465 but its graphics memory is about as fast as that of the GeForce GTX 470. By the way, one of the unique features of the memory subsystem of the new-generation Quadro series which distinguishes them from their gaming counterparts is that they support ECC.
The Quadro 5000 itself looks ordinary enough. Like the gaming GF100-based solutions, it is equipped with a massive cooling system that uses heat pipes and occupies two slots. Its plastic casing is silvery.
The Quadro 5000 seems to have an original PCB design with memory chips located on both sides, so there is no wonder that the reverse side of the PCB is almost fully covered by a heatsink plate.
The Quadro 5000 is about 2 centimeters shorter than the Quadro FX 4800 and will fit into the same system cases easily. It is about 1 centimeter longer than the GeForce GTX 465, though.
The Quadro 5000’s cooling system seems to be less efficient than the Quadro FX 4800’s as it has a smaller fan. This doesn’t make it loud, though. Its noise is acceptable even at high loads, which must be due to rather low heat dissipation of the GPU. The latter fact is confirmed by the card’s having only one 6-pin connector for additional power supply. According to the official specs, the new card has a peak power consumption of 156 watts, which is 6 watts higher than that of the Quadro FX 4800.
The new professional card has the same connectors as its predecessor: one dual-link DVI connector that supports LCD monitors with a resolution up to 3840x2400 pixels and two DisplayPorts.
DisplayPort has been given preference over the more widespread HDMI interface because it can transmit color information in 30-bit format which is supported by the Fermi-based Quadro cards. As a result, the card can display up to 1.7 billion colors if used together with a Deep Color monitor, e.g. an HP DreamColor LP2480zx, whereas standard solutions can only yield 16.7 million colors.
A standard 3-pin VESA socket for stereo glasses can be found near the video interfaces. This needs a special mention since Nvidia has begun to ship professional glasses 3D Vision Pro along with the new Quadro series. The Pro glasses differ from the consumer version in using a radio interface rather than an infrared channel for transmitting control signals. This expands the coverage and makes it possible to use multiple transmitters connected to different workstations within a single room.
There are two onboard connectors: SLI and SDI.
The former is for combining multiple professional cards into a single graphics subsystem for higher performance, more advanced antialiasing modes and quad-monitor configurations. The SDI connector can be used to connect a daughter card that enhances the main card’s functionality so that the Quadro 5000 could be used in digital broadcastinng.
Now that we’ve learned something about the Quadro 5000, we can read through its specs in comparison with the Quadro FX 4800:
It is not only the hardware capabilities of professional graphics cards that ensure high performance in CAD and CAM applications because optimized drivers are an important factor, too. Nvidia supplies special software for its Quadro series which differs from its ordinary gaming card drivers. It must be noted that, despite the similar hardware, the drivers for the Quadro and GeForce series cards are not compatible.
The professional and gaming drivers have a lot in common, though. For example, they have the same interface of the control panel.
Most of the settings are identical, to. In fact, the Quadro series drivers have the same basic options but feature setup profiles optimized for professional 3D applications rather than games.
There are other differences, though. As opposed to gaming cards, even based on the GF100 too, the Quadro 5000 supports higher antialiasing modes, up to 32x. If you combine two Quadro 5000 cards in SLI mode, you will be able to use even higher antialiasing modes, up to 128x.
Besides, the Quadro series drivers have built-in support for the stereo glasses and do not require that you install some additional software as you do with the GeForce series. The stereo technology itself differs somewhat on the Quadro series as the professional card supports stereoscopic vision not only in full-screen but also in windowed mode.
There are also unique options in the control panel of the Quadro driver. For example, you can enable ECC for the graphics memory.
Besides optimizing the basic drivers for CAD and CAM applications, Nvidia provides specialized mini drivers for graphics acceleration in Autodesk 3ds max and Autodesk AutoCAD. The mini drivers provide a substantial performance boost in these applications.
We tested professional graphics cards on a workstation built on the today’s fastest six-core processor – Intel Core i7-980X Extreme Edition working at 3.33 GHz frequency. Besides the CPU we also used a mainboard on Intel X58 Express and 6 GB of high-speed DDR3-1600 SDRAM.
In this test session Quadro 5000 will compete against the previous generation professional graphics accelerator from Nvidia – Quadro FX 4800. Moreover, we also tested one of the gaming graphics cards with the same architecture as Quadro 5000 in order to illustrate the differences between professional and consumer solutions.
Unfortunately, we didn’t get AMD’s FirePro professional graphics accelerators at the time of tests, so they won’t be included into this session. However, it is no big deal, in this case. Firstly, AMD solutions are not very popular in the professional graphics cards market. And secondly, within the next few months we are going to complete a massive comparative test session of all contemporary professional graphics solutions, which will of course include FirePro products, too.
So, for our today’s tests we put together the following testbed with the following software and drivers:
We tested our graphics cards in 1920x1200 resolution with Vsync disabled. We used popular CAD and 3D modeling applications and Standard Performance Evaluation Corporation (SPEC) benchmarks. The diagrams below present test scores in points, so the higher value indicates better performance.
The synthetic benchmark SPECViewperf is the first tool we use when testing professional graphics cards as it has become an industry standard for evaluating top-end graphics workstations. Modeling rather primitive operations, this benchmark shows the geometrical performance of graphics cards working via OpenGL. Such performance depends on both hardware resources and driver optimizations. The benchmark works by sending predefined sequences of OpenGL instructions to the graphics driver to visualize rotating models in the same ways as in certain professional applications.
A new version of SPECViewperf has been released recently. It features a revised list of tests and more complex 3D models. The scripts of the new SPECViewperf simulate a user working in the projection windows of the following professional applications (the name of each test is given in brackets): LightWave (light01), CATIA (catia-03), EnSight (ensight-04), Maya (maya-03), Pro/ENGINEER (proe-05), SolidWorks (sw-03), Siemens Teamcenter Visualization Mockup (tcvis-02), and Siemens NX (snx-01).
So, the new Quadro 5000 is considerably faster than its previous-generation predecessor. Talking about the superiority of the new Fermi-based Quadro series, Nvidia’s marketing department uses the word “exponential”. Indeed, the new card is more than twice as fast as the Quadro FX 4800 in such tests as ensight-04, tcvis-02 and snx-01. On the other hand, the Quadro 5000 has no advantage in light01 and proe-05. So, certain professional applications may expose some weaknesses in the overall more progressive Fermi architecture.
SPECViewperf also makes it clear that using gaming cards for professional applications is hardly appropriate. Although the GeForce GTX 470 has the same architecture as the Quadro 5000 and even works at higher frequencies, it has problems processing complex models via OpenGL. The speed of the gaming card is much lower than that of the specialized solutions with optimized drivers.
We also used SPECViewperf 11.0 to see how full-screen antialiasing affects graphics performance. The diagram below shows the changes in the SPECViewperf results when we enable different FSAA modes.
The large amount of graphics memory on board the Quadro 5000 helps this card support higher FSAA modes. It is the only one to support 64x mode whereas its predecessor Quadro FX 4800 was limited to 32x. Such FSAA modes provoke a considerable performance hit, though. On the other hand, the Fermi-based professional graphics card can boast higher memory bandwidth and thus enjoys an even larger advantage when we enable FSAA.
SPECViewperf 11.0 provides a good overall notion of the performance of the new professional graphics card, but it is a synthetic benchmark after all. Therefore, we will additionally test the Quadro 5000 in some real applications.
Unfortunately, the 3ds max benchmark from SPEC only works with the ninth version of the popular 3D modeling suite released back in 2006. However, this version supports hardware shaders and accelerated rendering of projection windows via Direct3D, which makes the results quite informative even today. This test is also interesting because Nvidia puts an emphasis on making its graphics cards fast in this popular 3D modeling suite and offers a special miniport driver, downloadable from its website, which integrates into the 3ds max suite and improves the speed of operations in the projection windows.
The new card is about 6% faster than its predecessor overall. That’s not as big a gap as we saw in SPECViewperf, but we shouldn’t neglect it, especially as the Quadro 5000 can be even faster in some special cases, e.g. when working with complex models that use both shadowing and wireframe elements. Besides, the small difference in the results is not only because the graphics cards are similar but rather because the application is too CPU-dependent to make full use of the graphics card’s resources.
Like 3ds max, Autodesk AutoCAD currently prefers the DirectX API. Moreover, the new versions of this computer-aided design suite do not even include a standard OpenGL driver. On the other hand, Nvidia offers its own nonstandard miniport driver for AutoCAD. This driver can use hardware shaders for rendering, so Nvidia’s professional cards show high performance. The Quadro series used to be head above their AMD opponents in our earlier AutoCAD tests. The new Quadro generation is going to push the bar even higher.
So, the new Quadro 5000 is 50% faster than its predecessor Quadro FX 4800 in AutoCAD 2011, which is a splendid result!
CINEBENCH R11.5 is a standard test of computer performance in the popular 3D modeling and animation suite Maxon CINEMA 4D R11.5.
Although the Quadro 5000 is ahead of the Quadro FX 4800 in this test, the difference is a mere 5%. This is due to the fact that CINEBENCH operates with textured 3D models while the speed of texturing is a weak point of the Fermi architecture.
There are no specialized mini-drivers for Maya 2011 from Nvidia, so this 3D suite uses the standard OpenGL API. Anyway, even the previous Quadro FX generation used to show very good results in this application, and the new Fermi-based models is going to improve them.
The Quadro 5000 is much faster than its predecessor in Maya 2011. The difference is 22% in the graphics component of the test.
SolidWorks produces unexpected results: the new Quadro 5000 is somewhat slower than the Quadro FX 4800. So, we can see the new card lagging behind not only in the synthetic SPECViewperf but also in real applications. We guess this is the result of the limited number of texture-mapping processors implemented in the Fermi architecture. There are only 44 of them in the Quadro 5000 whereas the Quadro FX 4800 has almost half as many as that!
Although professional graphics cards are hardly ever used for gaming, we want to benchmark them in the popular gaming test Futuremark 3DMark Vantage to take a look at them from this aspect.
Gaming load differs greatly from what the graphics card has to face in professional 3D applications. There is no wonder that the GeForce GTX 470 is much faster than the Quadro 5000. The GF100 chip of the gaming card has more unlocked multiprocessors and thus delivers higher performance in games where the speed of shaders is more important than the speed of drawing and texturing of triangles. The diagram above makes it clear that proper driver optimizations can do wonders and that the hardware resources of a graphics card are but one of several factors that affect its performance in professional design and engineering applications.
Now we will measure how much power workstations with professional graphics cards need. According to our method, we measure the power consumption of all system components (excluding the monitor) after the power supply unit. Thus, the efficiency of the PSU does not affect the result.
Besides idle mode, we loaded the systems with FurMark 1.8.2 running in Burn mode in a 1280x1024 window. Like most professional applications, this test uses OpenGL and loads the graphics subsystem heavily.
Although the Quadro 5000 needs more power than the Quadro FX 4800 at high load, the difference is small. The new card’s cooler is not any louder and its GPU temperature is no higher than 90°C.
The Quadro 5000 looks very good in comparison with the GeForce GTX 470 which is equipped with two 6-pin power connectors. The larger number of active shader processors and the higher GPU frequency of the GTX 470 show up in this test, and the Quadro 5000 consumes 50 watts less when running FurMark.
Take note that the Quadro 5000 needs somewhat more power than the other cards in idle mode. The difference of 8 watts is small but has to be accounted for. This is the tradeoff for the large amount of graphics memory which needs powering up even when idle.
The new professional graphics card Quadro 5000 from Nvidia is a success. Using one GPU with Fermi architecture, it delivers higher performance than its predecessor Quadro FX 4800. We have not found any obvious defects in it, actually. The Quadro 5000 is not hotter or louder and does not consume much more power than its precursor but boasts 1 GB more of graphics memory for processing complex 3D models and supporting higher FSAA modes. Like the FX 4800, the Quadro 5000 supports SDI daughter cards and can work in SLI configurations. Most importantly, the Quadro 5000 comes at a recommended price of $2249, which is quite a normal price for a top-performance professional graphics card.
In other words, the GF100 chip, not blameless on gaming graphics cards, turns to be optimal for fast graphics workstations.
Although we couldn’t compare the Quadro 5000 with AMD’s new-generation professional cards in this review, we are sure that Nvidia’s offer is going to be more appealing for most engineers and designers as it comes with better manufacturer’s support. Nvidia develops specialized software, including mini-drivers for the most popular 3D suites, to improve performance of its solutions even more. So, we don’t expect any changes in the trends on the professional graphics card market: Nvidia seems to be ready to expand its market share further.
A comprehensive comparative review of new-generation professional graphics cards from both Nvidia and AMD is on our plans, so stay tuned with us!