by Ilya Gavrichenkov
09/02/2013 | 03:31 AM
Besides gaming graphics cards, we also test professional products in our labs. However, we have mostly limited ourselves to Nvidia solutions, leaving AMD ones out of our scope. For example, in our recent review of the Quadro K5000 we compared it with the previous generation of Quadro cards rather than with AMD’s FirePro W series.
This didn’t make our review inadequate, though. The fact is that AMD’s professional graphics solutions are far from popular. The company’s share of this market has varied from 10 to 20% in the last few years, being much smaller than Nvidia’s. And there are reasons for that. Professional users need a special attitude, but AMD didn’t engage into a dialogue with them, letting its graphics cards sell by themselves. This marketing policy only works with gaming products, however. Indeed, the Radeon HD series owes its popularity to AMD’s pricing policy, always offering an attractive price/performance ratio, but this approach is not efficient when it comes to graphics cards for high-performance workstations. Price is just not the main priority for professional users who run computer-aided design and modeling applications.
The key priorities are high performance in CAD/CAM systems and appropriate tech support. AMD has not been doing well in this respect. The company didn’t work closely with developers of professional software suites, which resulted in the lack of application-specific driver optimizations. So, there were problems with speed as well as with hardware certification. It is not surprising then that system integrators that supply professional workstations were generally far from optimistic about the FirePro series.
This situation is likely to change now. The PC market has been shrinking, making the graphics card developers think about other market niches and high-performance workstations look like a lucrative one. Even though sales volumes are much smaller there in comparison with the PC market, professional graphics cards are more expensive, so the resulting profits may be comparable to those from gaming products. That’s why AMD has been paying more attention to the professional market lately. The company has updated its product range, transferring it to the Southern Islands GPUs, and has begun to actively cooperate with CAD/CAM application developers (with PTC in the first place) and add application-specific optimizations into the drivers.
AMD’s marketing policy has changed as well. The company has become interested in independent testing of its new-generation FirePro solutions, and that’s how we’ve got a couple of AMD products we are going to study in this review.
There is another reason for our writing this review, though. Nvidia has recently extended its product range of new-generation professional products. As opposed to AMD which focuses on top-performance cards right now, Nvidia offers a full spectrum of hardware CAD/CAM solutions. The Quadro K5000 model is now accompanied by new Kepler-based cards some of which we’ve got for us to test, too.
Thus, we can carry out a comparative test of the following professional graphics cards: AMD FirePro W8000, AMD FirePro W7000, Nvidia Quadro K5000, Nvidia Quadro K4000, Nvidia Quadro K2000 and Nvidia Quadro K600. In other words, we have CAD/CAM-targeted solutions from both GPU developers and their prices range from $170 to $1800.
AMD’s FirePro W series cards feature GPUs with the Graphics Core Next architecture, effectively being professional modifications of the Radeon HD 7000 series. The FirePro W8000 is not the top-of-the-line model as AMD also offers the more expensive W9000 which costs about $3400. We couldn’t get the latter, but the FirePro W8000 is in fact the same card, just somewhat slower and much cheaper. Both are based on the Tahiti GPU, the W8000 having a cut-down version of the chip. In its full configuration the GPU has 32 execution modules and 2048 shader processors whereas the W8000’s GPU has 28 execution modules and 1792 shader processors.
The memory bus is narrowed from 384 to 256 bits on the FirePro W8000 which is equipped with 4 GB of onboard memory. The number of raster operators, which are separate from the memory controller in Southern Islands GPUs, is 32, just like in the W9000 model.
The FirePro W8000 is targeted at high-performance workstations that have certain limitations concerning the power consumption and heat dissipation of the graphics subsystem. That’s why the W8000 needs only 150 watts of power via its two 6-pin connectors. Besides disabling some of the GPU subunits, the clock rates were reduced to lower the power draw. The GPU is clocked at 900 MHz and the memory, at 5.5 GHz. The clock rates don’t differ much from the W9000’s: its GPU frequency is a mere 75 MHz higher and its memory frequency is the same. However, the overall effect is that the W8000 has considerably lower specs than its senior cousin. Its pixel and texture fill rates are 28.8 Gpixel/s and 100 Gtexel/s, respectively, which is 20% lower compared to the W9000. The memory bandwidth is lower by a third at 176 GB/s.
The lower specs ensure lower power consumption, so the FirePro W8000 needs up to 189 watts as compared to its senior cousin’s 274 watts. And even with such specs as it has, the FirePro W8000 features higher theoretical performance than its opponent Quadro K5000. The latter is more economical, though. It needs 122 watts and has only one power connector.
The FirePro W8000 has quite an impressive appearance. It is a full-size card with a dual-slot cooler borrowed from top-end gaming single-GPU products. The full length (with the cooler) is 28 cm.
Although the FirePro W8000 has the same cooler as the Radeon HD 7970 (with a copper vapor chamber and an improved fan), it looks unique. We don’t mean the words on its plastic casing. The key difference is that the W8000 has an additional heat-spreading backplate which protects it from physical damage and also cools the chips on the reverse side of the PCB.
The card is also original in terms of video interfaces. It lacks DVI connectors but has as many as four DisplayPorts. You can also connect a DVI monitor to your FirePro W8000 by means of an adapter, which is usually included into the product box. A 3-pin connector for stereoscopic functionality can be found nearby. There are also internal connectors on the PCB: two for the CrossFireX technology and one for the synchronization module FirePro S400 (it supports Framelock and Genlock operation modes).
The FirePro W8000 costs $1400. Like other professional solutions from AMD, it comes with a 3-year warranty, comprehensive tech support and guaranteed 3-year lifecycle.
The FirePro W7000 is a professional graphics card that’s entirely different from the top-end models of the FirePro series. It has a Southern Islands GPU, too, but it is the simpler Pitcairn rather than the Tahiti chip. In other words, the W8000 is related to the Radeon HD 7970 and HD 7950 whereas the W7000 is a professional counterpart of the Radeon HD 7870. So while the FirePro W9000 and W8000 models came to replace the FirePro V9800, the FirePro W7000 is a new solution of the FirePro V7900 class.
On the architectural level, the FirePro W7000 sports a full-featured Pitcairn GPU without any disabled subunits. Thus, the GPU has 20 execution modules with 1280 shader processors and 32 raster operators. Like the W8000, it has 4 GB of onboard memory connected via a 256-bit bus.
Notwithstanding certain simplifications, the FirePro W7000 has many things in common with the W8000. They offer different computing performance but are similar in terms of graphics memory. The clock rates are somewhat different, too. The FirePro W7000 clocks its GPU and memory at 950 MHz and 4.8 GHz, so its pixel fill rate is higher at 30.4 Gpixel/s than that of the W8000. Its texture fill rate is 76 Gtexel/s. The peak memory bandwidth is 153 GB/s.
Since the Pitcairn is overall simpler than the Tahiti, the FirePro W7000 looks very economical in comparison with AMD’s flagship professional graphics cards. Its peak power draw is specified to be lower than 150 watts, and its PCB design indicates this fact. It only has one 6-pin power connector and is equipped with a single-slot cooler.
The cooler is interesting for having no counterparts among AMD’s gaming products which generally come with compact, yet still dual-slot coolers. The FirePro W7000’s cooling system looks like a flattened cooler from the W8000, which has a negative effect on its noisiness. The fan has to rotate at rather high speeds on this card, making the W7000 the noisiest among the professional cards covered in this review. One more peculiarity of the W7000’s single-slot cooler is that it doesn’t exhaust the hot air out of the computer case. By the way, the W7000 is not much shorter at 24 cm than its senior cousin.
The FirePro W7000 has the same video interfaces as the W8000, i.e. four DisplayPorts. DVI monitors can be connected to it via an adapter. With MST hubs, you can connect up to six monitors, as to any other top-end professional cards from AMD. Moreover, the FirePro W7000 has a connector for building CrossFireX configurations and a connector for the synchronization module FirePro S400.
The FirePro W7000 costs about $750, which is only half the price of the W8000.
It is the second time that we cover the Quadro K5000 in our reviews. Earlier we wrote a special report about it. Since then Nvidia has announced a few more professional products with Kepler architecture, yet the K5000 has remained the topmost offer in the series. It is an interesting fact as the card is based on the GK104 rather than on the faster GK110 GPU that is employed on Tesla computing cards and flagship gaming products. Well, the GK104 has enough resources for professional 3D CAD/CAM applications too, especially as it works in its full configuration, without any disabled subunits, on the K5000.
To be specific, the Quadro K5000 has 1536 CUDA cores, 128 texture-mapping units, and 32 raster operators. It has 4 GB of memory connected via a 256-bit bus. Apart from the increased memory amount, the specs are identical to those of the GeForce GTX 680 gaming card but the professional K5000 has considerably lower clock rates. Its GPU is 30% slower at 706 MHz and its memory is 10% slower at 5.4 GHz. This is meant to make the professional card quieter and more economical. Indeed, the K5000 looks preferable to AMD solutions in this respect. Its peak power draw is specified to be 122 watts and it only has one 6-pin power connector.
The theoretical performance specs are high enough for a professional product. The fill rate is 22.5 Gpixel/s and the texture sampling rate is 90.4 Gtexel/s. The peak memory bandwidth is 173 GB/s. This is but slightly lower compared to the competing FirePro W8000 but the K5000 is far more energy efficient.
Despite its rather low power consumption and heat dissipation, the K5000 has a full-featured cooling system. While many other cards with higher power requirements may have a single-slot cooler, the K5000 has a dual-slot one. The cooler has heat pipes in its base and exhausts the hot air out of the computer case. Its design is taken from top-end gaming cards with much higher power consumption, so it doesn’t have to exert itself to cool the K5000. As a result, the card is exceedingly quiet but large. Its length is 27 cm, although its PCB is only half as long.
Like the gaming cards of the Kepler generation, the Quadro K5000 is equipped with four video outputs: two DisplayPorts, one dual-link DVI-I and one dual-link DVI-D. As opposed to AMD, Nvidia lets you connect monitors via DVI without active adapters. It only supports up to four monitors concurrently whereas AMD solutions, up to six.
The K5000 has two onboard connectors for multi-GPU configurations. Using Maximus technology, the K5000 can be combined in SLI mode with Kepler-based Tesla computing cards. The result would be an all-purpose workstation for GPGPU and 3D graphics applications. The K5000 also has an onboard SDI/G-Sync connector to support the Quadro Sync system for synchronizing video outputs of multiple graphics cards.
The Nvidia Quadro K5000 is priced at somewhat higher than $1600 right now and usually comes with a 3-year warranty.
Nvidia follows the same principles when forming its professional product range as with its gaming products, so some of the Quadro series cards have cut-down GPU versions. The Quadro K4000 is one example of that. It is based on the GK104 chip with Kepler architecture we saw on the K5000. But the K4000 is a midrange solution, so it has less of graphics memory and a narrower memory bus. The number of active GPU subunits is smaller, too.
In fact, the GK104 has a very limited configuration on the Quadro K4000, so this professional card has no gaming counterparts. There are only half the CUDA processors left: 768. The number of raster operators is reduced to 24. As a kind of compensation, the K4000 has a higher GPU clock rate than the K5000 – 811 MHz. With such specs as it has, the Quadro K4000 might as well be based on the simpler GK106 chip, but it is not for some reason. This may still be done in the future, though. In this case, the professional K4000 card would become similar to the gaming GTX 660 model.
Meanwhile, the dramatic differences in specs between the K5000 and K4000 shouldn’t mislead you. As a matter of fact, the majority of professional applications do not really need a high-performance shader domain, using only rasterization and texture-mapping units. It means that the actual performance of the K4000 in CAD applications is not going to differ much from the performance of the Quadro K5000. For example, the K4000’s theoretical pixel fill rate is 19.5 Gpixel/s, which is a mere 15% lower than that of the senior Kepler-based professional card.
The Quadro K4000 has a memory frequency of 5.6 GHz to ensure a peak memory bandwidth of 134.8 GB/s. That’s not high because of the rather narrow 192-bit memory bus. The K4000 is equipped with a generous 3 gigabytes of GDDR5 but, unlike the K5000, does not support ECC.
Although the Quadro K4000 may seem a very cut-down product in comparison with the K5000, its specs are optimal for the majority of professional applications. And the K4000 is more convenient in terms of system integration. It is more economical than its Fermi-based predecessor Quadro 4000, requiring no more than 80 watts (but it has an additional power connector anyway). It has a single-slot cooler which, considering the low heat dissipation of the K4000, is quite comfortable acoustically. The card is not long at only 24.5 cm.
The Nvidia Quadro K4000 has one DVI-I port and two DisplayPorts but doesn’t support multi-monitor Quadro Sync and multi-GPU SLI configurations. If you don’t need such advanced technologies but want a high-quality card for CAD/CAM applications, the Quadro K4000 offers all the attributes of a professional solution (3-year warranty, comprehensive tech support, software certifications) at an attractive price. The K4000 sells in retail for about $800, which is only half the price of the flagship model.
Of course, not all of Nvidia’s latest professional graphics cards are based on the GK104 chip. The company has cheaper GPUs, GK106 and GK107, which are used on Quadro series products, too. One of them is the Quadro K2000 model, which features the GK107 chip. The GPU works in its full configuration without any disabled subunits, which makes the Quadro K2000 the professional counterpart of the gaming GeForce GTX 650 model. Their GPUs incorporate 384 unified shader processors, 32 texture-mapping units and 16 raster operators. The K2000’s GPU clock rate is rather high at 954 MHz, so the fill rate is 15.3 Gpixel/s while the texture sampling rate is 30.5 Gtexel/s. That’s not so bad in comparison with the K4000, which is only 30% faster, at least theoretically. The Quadro K2000 is rather weak in the computing department, having a small number of CUDA processors, but, as we've mentioned above, it is hardly a serious downside for a professional card designed for 3D CAD/CAM applications.
The Nvidia Quadro K2000 is equipped with a standard 2 gigabytes of GDDR5 memory which is clocked at 5.0 GHz. Coupled with the 128-bit bus, the peak memory bandwidth is 64.0 GB/s.
Professional graphics cards try to avoid dual-slot coolers because graphics workstations usually have much more expansion cards than home computers. Compactness is valuable there. That’s why the Quadro K2000 is equipped with a flat and slim cooler. It uses a centrifugal blower, just like coolers of high-performance cards, but it is small because the GK107 chip has rather low heat dissipation. The cooler doesn’t even cover the entire PCB, so it cannot exhaust the hot air out of the computer case.
The cooler’s fan doesn’t accelerate to high speeds, making the card rather quiet. The K2000 is also economical with a peak power draw of only 51 watts, so it doesn’t have an additional power connector.
The Quadro K2000 has the same video outputs as the K4000: two DisplayPorts and one DVI-I. By the way, the new professional product series with Kepler architecture is overall better than their predecessors in terms of multi-monitor configurations. Even the K2000 supports up to four monitors simultaneously (via an MST hub) and stereo 3D functionality (via USB).
The Quadro K2000 comes at a rather low price of $430, offering a very appealing combination of performance, power consumption, and cost.
Launching its Kepler-based series of professional graphics cards, Nvidia released a new product into every price segment. So while AMD offers previous-generation products for entry-level graphics workstations, Nvidia has a special entry-level card K600. Based on the same GK107 chip as the Quadro K2000, the junior model is just incredibly small.
And that’s actually the second point in favor of the K600, after its low price. This graphics card is going to be perfect for professionals who need a certified CAD/CAM solution for a small-size graphics workstation. Being a low-profile card, the Quadro K600 can be easily installed into a compact computer case.
Power and cooling shouldn’t be a problem as the Quadro K600 needs no more than 41 watts. It gets all of this juice from the mainboard and doesn’t need any additional power cables. A small cooler with centrifugal blower is mounted on its GPU. It is a smaller copy of the coolers installed on the more advanced cards. Although the fan is 40 mm in diameter, it never gets noisy just because its speed is never really high.
The cooler copes easily with the 28nm GK107 chip which works on the Quadro K600 at a rather low clock rate and in a cut-down configuration. One of its two SMX multiprocessors is disabled, so the number of unified shader processors is limited to 192 and the number of texture-mapping units, to 16. The only thing the Quadro K600 has in common with the K2000, which is based on the same GPU, is the number of raster operators – 16.
The K600’s GPU is clocked at 876 MHz, so the theoretical fill rate is 14 Gpixel/s. The texture sampling rate is 14 Gtexel/s. The DDR3 memory is connected via a 128-bit bus and clocked at 1.8 GHz. The peak memory bandwidth is 28.5 GB/s.
The Quadro K600 carries more memory than Nvidia’s earlier entry-level professional products. 1 gigabyte looks like a big step forward compared to the Fermi-based generation. On the other hand, it may be not enough for hard work in professional 3D design applications, so you may want to consider other products if you often work with large and complex 3D models.
The Quadro K600 has some other limitations in its design. It has only two video outputs (DVI-I and DisplayPort) and supports just two monitors concurrently. On the other hand, it is compatible with the Mosaic technology, even though in a somewhat limited implementation.
Now, the most interesting thing about the Quadro K600 is its price. The card costs a mere $180. You can only get previous-generation products like FirePro V4900 or Quadro 600 for that money among graphics cards for CAD/CAM applications, but they have worse specs.
Before we proceed to our testing of professional graphics cards, we want to sum up their specs in a single table. It will help us analyze the results we are going to obtain.
The testbed we use for our testing of professional graphics cards includes an Intel Core i7-3970X Extreme Edition processor, which is the fastest desktop six-core CPU available and works at 3.5 GHz, an Intel X79 Express mainboard, and 16 gigabytes of DDR3-1867 SDRAM.
So, our testbeds were built with the following hardware and software components:
We test professional graphics cards at a resolution of 1920x1200 pixels with Vsync turned off in popular professional applications for computer-aided design and 3D modeling. We also use special benchmarks from Standard Performance Evaluation Corporation (SPEC). The diagrams below show the test scores in points, so a bigger value always means higher performance.
In our tests of professional graphics cards, we first run SPECViewperf, which has become an industry standard for benchmarking graphics workstations. SPECViewperf sends predefined sequences of OpenGL instructions to the graphics card driver, describing rotations of complex models typical of various professional applications. By simulating such rather primitive operations, this benchmark reports the geometrical performance of a graphics card which is determined by hardware features as well as driver optimizations.
The scripts integrated into SPECviewperf version 11 emulate the user’s work in wireframe mode in the following professional applications (the names of corresponding tests are given in brackets): LightWave (lightwave-01), 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).
Nvidia’s professional cards have higher results in the synthetic benchmarks than the AMD FirePro series. Judging by the SPECviewperf scores, the FirePro W8000 and W7000 are only comparable to the Quadro K2000, which doesn't agree well with their market positioning. We can also note that AMD’s cards from different price categories and with different GPUs have very similar results in quite a lot of test scenarios. This is due to the specifics of SPECviewperf which generates a stream of rather simple commands that make heavy use of the GPU's raster operators but hardly ever need the shader domain. Modern versions of CAD/CAM applications have moved away from such visualization mechanisms, trying to put the graphics card’s shader processors to a better use. That’s how AMD itself explains the rather poor performance of the new FirePro cards in SPECviewperf. In other words, this benchmark is not the final truth at all and cannot replace tests in real-life applications.
For all its downsides, SPECviewperf 11.0 can show us the effect of full-screen antialiasing on performance. In the diagrams below you can see how the SPECviewperf scores change when we enable various FSAA modes.
The professional cards from AMD and Nvidia differ in their behavior with full-screen antialiasing. Nvidia's cards slow down as the level of FSAA increases, which is just what we might expect. The AMD FirePro series, however, often deliver the same performance irrespective of FSAA levels. Moreover, they do not support FSAA levels higher than x16. All of this makes us suspect the FirePro W series driver to share a large amount of code with the gaming Catalyst driver. The FirePro series behave like gaming rather than professional cards in SPECviewperf.
So, while Nvidia implements general OpenGL optimizations in its driver for professional products, AMD seems to focus on performance in specific applications and borrows a lot of features from the gaming driver. We guess each approach is viable but you should be aware that FirePro cards may potentially have more problems with specialized OpenGL applications than the Quadro series.
While providing a good general notion of the professional cards’ OpenGL performance, SPECviewperf 11.0 is still just a synthetic benchmark. That’s why we don’t limit our testing to it. We’ll also run a few actual professional applications.
We used the professional version of the SPEC benchmark to check out the graphics cards’ performance in one of the most popular 3D modeling suites. Besides other things, that version is noted for employing extremely complex models with some 32 million polygons and evaluating performance separately for ordinary and complex models.
Nvidia and AMD optimize their professional solutions for 3ds Max projection windows by offering special mini-drivers that replace the standard DirectX driver in the software suite and ensure higher performance for the Quadro and FirePro series.
The graphics cards have the same standings in 3ds Max 2011 as in SPECviewperf. The FirePro series products, W8000 and W7000, deliver almost the same performance and fall behind all of the Kepler-based Quadro series cards, except for the junior model. Even the Quadro K2000 is easily ahead of the FirePro W7000 by as much as 43%. As for the Quadro K600, it couldn’t complete the test because 1 GB of onboard memory was not enough to process its complex 3D models. But the next card, Quadro K2000, beats the FirePro W8000 while costing only half the price of the FirePro W7000.
The standings change, however, when the professional cards have to deal with extremely complex models.
The FirePro W series unexpectedly get a second wind when they have to process a city model made of 32 million polygons. They become considerably faster than their Quadro series opponents. This interesting outcome is partly due to the specifics of the 3ds Max 2011 suite which uses DirectX for projection window visualizations instead of OpenGL which is generally more popular for professional applications.
It is the single usage scenario in 3ds Max 2011 where AMD’s professional cards can deliver high performance, though. Just take a look at the secondary performance indexes issued by the benchmark:
Alas, the FirePro W8000 and W7000 do not show anything spectacular here. Nvidia's Quadro series cards are still the better option for 3ds Max. They ensure generally higher performance in the projection windows and accelerate the rendering process far more efficiently than their competitors.
Here’s another popular 3D modeling suite that uses DirectX. In our previous reviews we could see that Nvidia’s professional solutions were no better than their gaming counterparts in this application. However, the situation has changed with the release of new drivers and the Quadro series has become considerably faster.
The AMD FirePro W8000 can offer the same performance in AutoCAD 2013 as the Quadro K5000. The AMD FirePro W7000 is comparable to the Quadro K2000. The AMD and Nvidia products differ in their behavior depending on the model representation mode.
We can see that AMD’s products are better in the simplest (Wireframe) and in the most complex (Realistic) mode. The Quadro series are preferable for working in the Hidden and Conceptual modes.
Maxon Cinema 4D is a popular software suite for 3D modeling and animation. Its speed can be evaluated by means of the specialized benchmark CINEBENCH. The latter uses the engine of Cinema 4D version R11.4 but, according to its developers, can also provide some notion of a graphics card’s performance in the up-to-date Cinema 4D R14.
There are as many as four professional cards that offer similar performance in this application. These are AMD’s FirePro W8000 and W7000 and Nvidia’s Quadro K5000 and K4000. It is quite possible that these flagship products are just so fast here that the workstation is limited by the performance of its CPU.
Creo is a popular CAD suite that has replaced Pro/Engineer. The benchmark we use features a car model that consists of a thousand components and needs about 1 gigabyte of graphics memory. It makes full use of the application’s functionality by enabling edge and reflection shading, hidden surfaces on/off, full-screen antialiasing (up to level x8), and photorealistic edge rendering.
In PTC Creo, the FirePro W8000 and FirePro W7000 are slower than the Quadro K5000 but faster than the Quadro K4000. However, we know that AMD collaborates closely with PTC, so Creo is one of the key applications AMD programmers optimize their driver for. As we can see, this doesn’t produce any miracles.
On the other hand, Creo Parametric 2.0 features a new visualization mode called Order Independent Transparency. This mode ensures the best reproduction of multiple overlapping transparent objects but is only currently supported on the FirePro series. That’s why it is not included into the formal performance benchmarks. So if you’re choosing a professional card for PTC Creo, you should be aware that AMD products can offer broader functionality.
The speed of processing models in other visualization modes is shown in the diagrams.
The competing cards from AMD and Nvidia deliver similar performance in the wireframe and shaded modes. Take note, however, that the FirePro W8000 and FirePro W7000 are very close to each other whereas the gaps between the adjacent models in the Kepler-based Quadro series vary from 10 to 50%.
The popular 3D graphics editor Maya 2013 is a typical professional application based on the OpenGL API. AMD's graphics cards used to be especially strong here just a few years ago, but the current situation is different.
AMD’s solutions W8000 and W7000 are in between Nvidia's Quadro K5000 and K4000 in terms of performance. This result is adequate for the FirePro W8000 considering its price, and very good for the W7000 which is slightly cheaper than the Quadro K4000. On the other hand, if we take the price factor into account, the junior Quadro series products must be noted. The K2000 and K600 offer the best price/performance ratio in this application. For example, the K2000 is only 30% slower than the flagship K5000 but costs many times less.
The standings of the AMD and Nvidia cards do not differ much in the various model visualization modes. Still, we can note that the Kepler-based Quadro series are especially preferable to their opponents when working with wireframe models.
NX is the flagship CAD/CAM/CAE PLM-system that is extensively used in engineering, so we just couldn’t help including it into our test session, especially as standings of the professional cards from AMD and Nvidia are different from what we’ve seen in most of the previous tests.
The AMD FirePro W series cards can offer much higher performance here than their Nvidia competitors. The W8000 is 37% faster than the Quadro K5000 whereas the W7000 is 35% faster than the Quadro K4000. The K4000 is for some reason slower than the K2000 which, considering its price, seems to be quite an attractive option for NX.
It’s even more interesting to check out the speed of the professional cards in different application modes rather than in terms of the overall performance indexes.
The FirePro series is fantastically fast in the wireframe mode: almost twice as fast as the Quadro series. Well, the FirePro W8000 is very good in the rest of the visualization modes, too. The FirePro W7000 is not as fast, only competing with the Quadro K2000. The Quadro K4000 is no good at all for Siemens NX: its performance in this application doesn't match its price.
SolidWorks is a highly popular CAD application. The new SPEC benchmark for this application can use a lot of features offered by modern graphics hardware. Its medium-complexity models take up to 1 gigabyte of graphics memory.
Like most other benchmarks, Solidworks 2013 puts the Quadro K5000 on top but the competing solutions from AMD perform well enough, especially considering their lower price. The FirePro W7000 is particularly good as it offers the same performance as the W8000 is quite a lot of tests. The difference between the W8000 and W7000 is only 3% in Solidworks, for example, but the senior model is twice as expensive, which makes the FirePro W7000 a highly interesting option, preferable to the Quadro K4000, which costs about the same money but is slower.
We can see the same standings in Solidworks irrespective of the model visualization mode.
There is only one exception. Turning off the Realview option lowers the load on the shader domain, so the FirePro series lose their ground. However, working in Solidworks without Realview is hardly interesting for owners of professional graphics cards because this visualization mode, unavailable on gaming cards, is the main reason for Solidworks users to switch to professional products.
Although professional graphics cards are almost never used for gaming, we still want to run the popular gaming benchmark Futuremark 3DMark Fire Strike to take a look at our graphics cards from a different angle.
Gaming is different from professional applications, as you can see. AMD cards are superior in 3D gaming performance to comparably priced Nvidia solutions but, as we've seen above, this is not true for professional applications. A graphics card’s hardware resources, especially its shader domain, are not the main factor that determines its performance in professional CAD/CAM applications. Driver optimizations and balanced specs are far more important. And, as you can see, Nvidia dominates the market of professional graphics cards by just being better than AMD at adapting its graphics architectures to CAD/CAM software.
In this section we want to show you how much power is needed by the complete systems (without the monitor) equipped with the tested graphics cards. We measure the power consumption by means of the Corsair AX1200i power supply. The result is the sum total of the consumption of each system component. The PSU’s efficiency doesn’t affect it.
There are two test modes: idle and high load (FurMark 1.9.2 running in Burn mode in a 1280x720 window). We use FurMark for the high-load mode since, like most professional applications, it is based on OpenGL and is a heavy load indeed.
Comparing professional cards from the same price categories, the Quadro series looks far more economical. The Quadro K5000 configuration needs almost 50 watts less than its FirePro W8000 counterpart. The Quadro K4000 consumes 34 watts less than the FirePro W7000. Nvidia puts an emphasis on this fact, counting the energy efficiency of its Quadro series among its advantages. Well, it might be expected as the GPUs of the Quadro cards do not have maximum specs and the cards themselves don’t have two additional power connectors.
The FirePro series, on its part, is close to gaming cards in its design. That’s why their heat dissipation and power consumption are comparable to those of the Radeon HD series, which means that AMD-based workstations are going to be noisier and hotter than their Nvidia-based counterparts.
There is a reason for Nvidia’s domination on the market of professional graphics cards. As we have made sure in our today’s tests, this manufacturer can adapt its graphics architectures well to the needs of 3D CAD/CAM application users. By optimizing both the graphics hardware and the driver, Nvidia offers a well-balanced series of professional cards that provide different levels of performance for different money.
AMD doesn’t have such a comprehensive strategy as yet. We couldn’t get the midrange FirePro W5000 model, but the FirePro W7000 and W8000 leave us rather perplexed. Their prices differ by almost 100%, yet they often show similar results in real-life professional applications. So, the FirePro W8000, which is supposed to compete with the Quadro K5000 according to its pricing, fails to do so. The FirePro W7000, on the contrary, is often faster than the comparably priced Quadro K4000, which makes the AMD product preferable as a midrange professional graphics card, especially for such applications as Creo, NX or Solidworks. If you want higher performance and choose from top-end products, the Quadro K5000 will be the best choice almost everywhere. The FirePro W8000 should only be considered for specific situations, for example for processing complex models in 3ds Max and NX.
Of course, AMD should be given credit for trying to improve its standing on the professional graphics card market. The company seems to be going in the right direction. The transition of the FirePro series to the GCN architecture and the close collaboration with developers of professional applications are highly positive trends. However, it is going to take more effort on AMD’s part to promote its FirePro products for graphics workstations of various levels. The overall OpenGL optimization of AMD’s professional graphics driver is far from perfect yet as we could see in our synthetic benchmarks. Moreover, AMD pays little attention to the heat dissipation and power consumption of its cards, so the FirePro series are inferior to their Quadro opponents in this respect.
One more advantage of Nvidia’s strategy is that the company’s model range includes inexpensive professional cards based on the modern Kepler architecture. They support all the latest graphics technologies. While AMD offers marked-down graphics cards of previous generations in the low price segment, Nvidia releases completely new solutions like the Quadro K2000 or Quadro K600 which offer a highly attractive price/performance ratio.
Summing everything up, we should say that Nvidia’s Quadro series of the 2013 model year, which features the modern Kepler architecture, remains the preferable choice for professional graphics workstations. The FirePro W series that AMD wants to improve its market standing with is only good in the middle price segment. We mean the FirePro W7000 model specifically. This graphics card offers high performance at an average price and looks good in a number of CAD/CAM applications, making it a worthy alternative to the Quadro K4000. It is clear, however, that one graphics card model is not enough for a good fight for the whole market.