by Alexey Stepin
04/05/2006 | 02:26 PM
January 24, the desktop 3D graphics market welcomed yet another landmark product, ATI Technologies’ Radeon X1900. The company claimed technological superiority with this chip that incorporated an unprecedented 48 pixel shader processors. Graphics cards carrying this GPU on board proved to deliver generally more performance than the GeForce 7800 GTX 512, not to mention Nvidia’s weaker products on the G70 graphics processor.
<%BANNER[article]%>A severe deficit of G70 chips capable of working at high clock rates transformed the single competitive solution from Nvidia in the ultra high-end sector – GeForce 7800 GTX 512 – into a rarity. Nvidia found itself virtually unarmed against the onslaught of ATI’s new Radeon X1900 series, although their next GPU was indeed under development. And on March 9, the GeForce 7900 graphics processor, codenamed G71, came out to succeed to the glorious G70.
Two new graphics card models were announced alongside: the GeForce 7900 GTX and GeForce 7900 GT are meant to replace the corresponding GeForce 7800 models and come at a recommended price of $499 and $299, respectively, which is much cheaper than the recommended price of the Radeon X1900 XTX ($549) and Radeon X1900 XT ($475). So, this is a very aggressive answer to ATI Technologies, yet it’s not enough just to announce a product to win the customer’s heart. The new product must be available for purchase and we hope Nvidia will do better now than with the GeForce 7800 GTX 512 and will provide the new cards in as many quantities as necessary.
And now let’s see what’s so new about the new graphics processor.
We talked about the ways to improve a graphics processor performance in our review of the Radeon X1900 architecture. Here are the main points for you, in brief.
So, one way is to increase the number of computational units that are responsible for pixel shader execution, the so-called pixel shader processors, sacrificing other subunits of the GPU like texture-mapping units (TMUs) or raster operators (ROPs). ATI Technologies went this way as they created their Radeon X1900 (R580). This approach looks justifiable considering the tendency for more math1ematical instructions in pixel shaders in contemporary and upcoming games, but a GPU of this design won’t do too well under high textural loads. You could see that in Radeon X1600 which was severely limited by its four TMUs in many of our tests (for details see our review called ATI RADEON X1600 XT: Mainstream Performance Redefined Once Again?). In the same manner, the new Radeon X1900 XTX was not any better than the older Radeon X1800 XT in certain tests because the two cards had the same number of TMUs and ROPs.
The way chosen by Nvidia implies that math1ematical computations and texture operations have the same share in games. A GPU created according to this principle has about the same number of pixel processors and TMUs. For example, the GeForce 7800 had 24 pixel processors, 24 TMUs and 16 ROPs and it felt at ease in textures-heavy applications. The new graphics processor from Nvidia carries this principle on, but some expectations concerning the G71 haven’t really come true. The G71 was expected to come out with 32 pixel processors, 32 TMUs and 16 ROPs, but Nvidia must have thought it proper not to redesign the G70 that dramatically, and the G71 differs from it but very little:
The table tells you the technical characteristics of the senior model of the new GeForce family, but the less powerful GeForce 7900 GT just has different clock rates, 450/660 (1320) MHz and is equipped with half the amount of graphics memory (256MB).
By the way, the G71 is the second desktop graphics processor to work at as high a frequency as 650MHz, ATI’s R580 (Radeon X1900 XTX) having been the first. Well, the entry-level Chrome S27 processor from S3 Graphics works at 700MHz but it is obviously a much simpler chip than the top-end solutions from ATI and Nvidia. So, although Nvidia cannot claim it produces higher-frequency chips, it has got even with ATI in this respect. We’ll see in the gaming tests section how the higher operating frequencies affect the performance of Nvidia’s new flagship product.
It seems the G71 differs from its predecessor in clock rates and tech process only, but take note of the dramatically smaller number of transistors the chip is made up of. The GeForce 7900 is 24 million transistors “thinner” than the GeForce 7800! This makes it the simplest graphics processor in the top-end sector today (fewer transistors were only employed in the previous-generation GeForce 6800 (NV40/45) and Radeon X800/850 (R420/R423/R480) processors). Combined with the thinner tech process, the fewer transistors mean a smaller die area and better thermal characteristics of the chip. The manufacturing cost is reduced, too, since more dies can be made out of a single silicon wafer.
Unfortunately, we do not know exactly how Nvidia managed to strip the chip of so many transistors without losing anything in its functionality. We do know that this was made by redesigning the pixel processors area of the chip. 24 million transistors is quite a big amount, so we are left to wonder what was cut off from the new GPU. The G70 may have originally been redundant or perhaps Nvidia did manage to optimize the pixel processors design so marvelously.
The chip now also incorporates two dual-link DVI controllers which have previously been available in Radeon X1000 series GPUs only. So, Nvidia’s new graphics cards do not need an additional chip to implement a dual-link DVI interface which is used with high-resolution TFT displays like the 30” Apple Cinema HD. CRT displays are supported exactly as before: two 400MHz RAMDACs allow using display modes up to 2048x1536@85Hz. It is unnecessary to improve anything in this area since CRT monitors are currently leaving the consumer market.
Nvidia returns in the GeForce 7900 to the idea of clocking different GPU subunits at different frequencies: the pixel processors and ROPs of the GeForce 7900 GTX and GT work at 650 and 450MHz, respectively, and the vertex processors at 700 and 470MHz, respectively. Considering that the speed of modern games is not generally limited by the speed of the vertex processors, this approach shouldn’t tell negatively on the performance of the new graphics cards.
This shouldn’t affect their thermal and electrical characteristics much, either. Nvidia declares a peak power consumption of 120W for the GeForce 7900, but the chip consumes that much power only when all its transistors are working simultaneously. This has nothing to do with real operating conditions, of course. Considering the smaller amount of transistors and the use of a new tech process, we can expect the GeForce 7900 GTX to be more economical than the GeForce 7800 GTX 512. We’ll check this out shortly, but now let’s take a closer look at the new card.
The senior model of the new family, GeForce 7900 GTX, uses the same PCB design as the GeForce 7800 GTX 512:
The two graphics cards have the same appearance and are equipped with identical coolers. You can only spot a few differences on the reverse side of the PCB unless you remove the cooler: a few chips and small elements are missing on the GeForce 7900 GTX near the power connector. They used to be part of the power circuit, obviously. When the cooler is removed, the much smaller GPU die and the empty seat previously occupied by a VIVO-supporting Philips SAA715HL chip catch your eyes. On a closer inspection, you can also note that some elements of the power circuit are missing here but were present on the GeForce 7800 GTX 512. This again proves that the G71 has lower power requirements than the G70 clocked at 550MHz.
There’s a huge difference between the die area of the G70 and G71 chips. The core of the new chip is almost 1.5 times smaller than of the older one. Even the thinner, 0.09-micron tech process couldn’t have helped to do that hadn’t Nvidia optimized the new GPU and made it slimmer by 24 million transistors. The G71 sample on our graphics card was manufactured on the 6th week of 2006, i.e. from February 6 to 12 or near two months ago. This is the third, A2 revision of the chip (counting from A0).
There is no protective frame around the die, although the smaller die is easier to damage. Nvidia’s cooling system isn’t too heavy, however, so the GPU will come to no harm if you handle the device carefully enough. Exactly complying with the specification, the graphics processor works at 650MHz. This frequency is the clock rate of the pixel processors and raster operators whereas the vertex processors of the GeForce 7900 GTX work at a higher clock rate of 700MHz.
Like the GeForce 7800 GTX 512, the GeForce 7900 GTX uses graphics memory from Samsung’s K4J52324QC series in the modern 136-pin FBGA packaging. The BJ-11 prefix indicates that the chips work at 2.0V voltage and have an access time of 1.1 nanoseconds which means operating frequencies up to 900 (1800) MHz. Each chip has a capacity of 512Mb and is organized as 16Mbx32, so eight such chips suffice to provide 512 megabytes of memory with a 256-bit memory bus. The memory is clocked at 800 (1600) MHz while the memory of the GeForce 7800 GTX 512 worked at 850 (1700) MHz. Thus, the peak memory bandwidth of the new card is lower in comparison with the previous model: 51.2GB/s against 54.4GB/s.
Nvidia first implemented this cooling solution in its professional graphics card Quadro FX 4500. Quiet and efficient, it first moved on to the GeForce 7800 GTX 512 (for details see our article called NVIDIA GeForce 7800 GTX 512: Faster, Higher, Stronger!) and then landed on the new GeForce 7900 GTX.
It’s designed quite simply: the copper sole that has direct contact with the GPU transfers its heat to the four heat pipes which pass the heat over to the two sets of thin aluminum ribs on the right and left of the fan; the ribs are covered with a profiled case. Only part of the hot air is exhausted outside, but practice suggests that this doesn’t affect the overall cooling efficiency.
The cooler uses an 80mm fan with good aerodynamics of the blades, which ensure high performance at a minimum noise. In fact, the cooler is practically silent since the fan speed is only about 1000rpm most of the time.
The cooling system is manufactured well. The aluminum ribs are pressed tightly to the heat pipes to ensure proper thermal contact. Both sections of the heatsink become very hot at work, indicting good heat transfer. Nvidia made the cooler light by using aluminum in it; there is also a smaller chance of damaging the graphics processor die under such a cooler.
Nvidia’s traditional thermal interfaces are employed here: a dark-gray thick thermal paste between the GPU and the cooler and soft pads soaked in white thermal paste between the memory chips and the cooler’s base.
This cooler has done well on the GeForce 7800 GTX 512. The GeForce 7900 GTX is expected to generate even less heat, so there should be no cooling-related problems at all with this graphics card.
Power consumption is a very important parameter of a modern graphics card, so each new device we receive in our labs has to pass this test. The appetites of Nvidia’s GeForce 7900 GTX and 7900 GT were measured on a special testbed with the following configuration:
We measured the power consumption of the cards with a digital multimeter Velleman DVM850BL (0.5% measurement accuracy). To put a Peak 3D load on the card we ran the first SM 3.0 graphics test from 3DMark06 in a loop at 1600x1200 resolution and with enabled 16x anisotropic filtering. Then we created an extremely high 2D load by launching the 2D Transparent Windows test from Futuremark PCMark05. Here are the results:
So, we’ve got a new record here! The new graphics card is more economical than the ex-leader GeForce 7800 GTX 512. The difference is 10 watts, which is an excellent result considering the higher clock rate of the GeForce 7900 GTX graphics processor. The power consumption of the new flagship GPU from Nvidia is in fact comparable to that of the GeForce 7800 GTX that has lower operating frequencies and less memory on board.
Note that the new card consumes the largest share, about 45W, from the PCI Express x16 slot and only about 36W from the external power line. The remaining 3W are consumed from the +3.3V line. For comparison, the Radeon X1900 XTX is almost the same load on the graphical slot, but it sucks in as much as 72W from the external +12V line! Thus, though an elite graphics card, the Nvidia GeForce 7900 GTX has a very modest appetite and doesn’t require a high-wattage power supply even if you are going to put two such cards into your system case. A good-quality 400-450W PSU will be quite enough.
The GeForce 7900 GT did well in the power consumption test, too. Having somewhat better parameters than the GeForce 7800 GTX, it consumes less power – the difference is over 30 watts! The 7900 GT is also more economical than the GeForce 7800 GT which has only 20 active pixel processors and a memory frequency of only 500 (1000) MHz. In fact, the GeForce 7900 GT consumes as much power as the GeForce 6600 GT that has only 8 pixel and 3 vertex processors! This graphics card could even do without any additional power at all since the PCI Express slot is capable of providing up to 75 watts to the installed device.
The load distribution is similar to what we saw with the GeForce 7800 GT. The GeForce 7900 GT consumes a mere 0.8W from the +3.3V line. The +12V load is evenly shared between the internal and external lines: 24 and 23 watts, respectively.
We guess there’s no need to talk about a minimum PSU wattage. Any high-quality power supply with stable output voltages will do. A SLI configuration with two GeForce 7900 GT is going to consume less than 100 watts of power, delivering a very high performance. This is an admirable achievement for Nvidia whose solutions have traditionally been more economical than ATI’s, at least in the top-end products sector.
The same cooler as on the GeForce 7800 GTX 512 is mounted on the GeForce 7900 GTX and it works as quietly, too. You can hardly hear the graphics card even with your ear close to the working computer. We don’t think you won’t want to replace the cooler even if perfect silence is your main priority. For comparison, the Radeon X1900 XTX produces some noise even when its blower is working at a reduced speed, and when the speed is high, the noise is hardly bearable due to the irritating tone produced by the resonating plastic casing.
It’s worse with the noise characteristics of the GeForce 7900 GT. The small blower this graphics card’s cooler is equipped with is always working at a high speed and is always audible against the noise from other computer components (CPU, power supply, hard disk drives). We didn’t find any traces of a speed management system on our sample of the card, but we think off-the-shelf samples of the GeForce 7900 GT will have it. The fan speed can be controlled with the latest version of RivaTuner, but we wouldn’t recommend you to do so since the GeForce 7900 GT cooler is small and may not cope with its job if the fan speed is set too low.
Franking speaking, we don’t think the GeForce 7900 GT cooler is half as good as the GeForce 7900 GTX one. It is small and has a tiny fan, so it is rather noisy, yet not very efficient. If you are going to overclock your GeForce 7900 GT or if you just want to have a silent computer, you may want to think about a more efficient cooling solution. It’s possible some manufacturers will equip their GeForce 7900 GT with such coolers right away.
An attempt to overclock our GeForce 7900 GTX brought modest results: the GPU clock rate grew up from 650 to 680MHz. We used the standard overclocking options available in the ForceWare driver, and we suspect the vertex processors of the chip were the limiting factor as they had to work at 730MHz. The memory was stable at 920 (1840) MHz or a little higher than it is rated for. Well, even this modest frequency growth is an achievement for chips that are originally clocked at their limit. We might have been luckier using water or cryogen cooling, but this kind of overclocking is generally costly and few people are actually engaged in it.
The GeForce 7900 GT is manufactured out of G71 chips that couldn’t pass the frequency check for GeForce 7900 GTX, so we hadn’t expected we would raise the core frequency of our sample of that card to 600MHz, not to mention 650MHz. That was indeed true. With its stock cooler and with a little help from an additional 120mm fan the card was stable at 560MHz core frequency and 800 (1600) MHz memory frequency. This is quite a good result for the humble cooler the GeForce 7900 GT is equipped with. Of course, the overclocking potential of different samples of that card is going to vary somewhat, yet we think an overclocker may be interested anyway considering the recommended price of $299.
As for 2D image quality, the new graphics cards from Nvidia produced a crystal-sharp image in every mode supported by our monitors, up to 1800x1440@75Hz, like most other modern graphics cards do.
Since we had two GeForce 7900 GTX and two GeForce 7900 GT graphics cards at our disposal, we decided to also investigate the performance of the corresponding SLI-tandems and compare the results against those of the ATI Radeon X1900 XT CrossFire. We used the following mainboards as a basis for our test platforms:
The remaining components were the same for both test platforms:
These games and applications were used as performance benchmarks:
First-Person 3D Shooters
Third-Person 3D Shooters
Simulators
Strategies
Semi-synthetic benchmarks
Synthetic benchmarks
We select the highest graphics quality settings in each game, the same for ATI’s and Nvidia’s solutions, except for the Pacific Fighters flight sim which requires vertex texturing support to enable its Shader Model 3.0 mode. The Radeon X1000 family doesn’t support this feature and runs the game in the Shader Model 2.0 mode. We do not edit the games’ configuration files. If possible, we use the games’ built-in benchmarking tools and if not, we measure the frame rate with the FRAPS utility. We measure minimal as well as average fps rates whenever possible.
To load the video subsystem to the full extent and to minimize the influence of the CPU speed on the performance results we didn’t test the systems in the “pure speed” mode. Instead, we added two new modes to our traditional “eye candy” settings (4x FSAA + 16x anisotropic filtering). These were the resource hungry aliasing modes, such as Super AA/SLI AA 8x + AF 16x and Super AA 14x/SLI AA 16x + AF 16x. We turned on the 4x FSAA + 16x AF mode from the game’s own menu if it was possible. Otherwise, we forced the necessary mode from the ForceWare driver as we also did for the higher levels of full-screen antialiasing.
The drivers were set up as usual.
ATI Catalyst :
Nvidia ForceWare:
Before we move on to the benchmark results in games, let’s discuss the results of our theoretical performance tests that were performed with Marco Dolenc’s Fillrate Tester and Xbitmark 0.65.
Having a higher GPU clock rate, the GeForce 7900 GTX has a higher scene fill rate than the GeForce 7800 GTX 512. We can’t see any anomalies in the new card’s behavior even though the new GPU has got slimmer by 25 million transistors in comparison with the older one. Nvidia’s solution now boasts a higher fill rate than the Radeon X1900 XTX even without texture mapping.
The same is true for the second test which measures the pixel processors performance. The GeForce 7900 GTX behaves identically to the GeForce 7800 GTX 512, the GPU frequency accounting for the difference in their results.
Xbitmark provides us with more detailed results. You can see that the core frequency increase from 550 to 650MHz has a very positive effect on the performance of the GeForce 7900 GTX at executing various pixel shaders. Having two times fewer pixel processors, Nvidia’s new GPU is at least no worse than the Radeon X1900 XTX even when running math1ematics-heavy shaders like Dot Product Bump Mapping + Specular + Reflection, Factored BRDF + HDR or Wood. Some of the pixel processors of the Radeon X1900 XTX must be just idle in Xbitmark because the scenes of this test use no more than 3-5 shaders at a time. As we know already, a GPU with a lot of pixel processors can only show its best when there are multiple math1ematics-heavy shaders to be performed simultaneously.
Having 24 TMUs, the GeForce 7900 GTX enjoys a bigger advantage in tests with pixel shaders that include multiple texture lookups (both the NPR Hatch shaders and Cook-Torrance + Freshnel). The only area ATI’s architecture remains unbeaten in is execution of shaders with dynamic branching. The Radeon X1900 XTX still has no rivals there.
All in all, the GeForce 7900 GTX does somewhat better than the Radeon X1900 XTX. The two graphics cards are roughly equivalent when it comes to executing pixel shaders, but Nvidia’s new solution is better than the senior Radeon at processing textures. The latter is indeed better on shaders with dynamic branching, but such shaders aren’t yet used in games.
And now it’s time to see how well the new family of graphics cards from Nvidia is going to do in real-life conditions that often have very little to do with theory.
Battlefield 2 still does not support extreme levels of full-screen antialiasing, so we limited ourselves to the 4x FSAA + 16x AF mode.
The GeForce 7900 GTX is as fast as the Radeon X1900 XTX in the two lower resolutions, but loses in the resolution of 1600x1200 – the less advanced memory controller is to blame, we think. It is also a mere 5fps ahead of the GeForce 7800 GTX 512 in 1600x1200, which points at the memory subsystem performance as the main bottleneck in this resolution because the gap is larger in the lower display modes.
The SLI configuration with two GeForce 7900 GTX is, on the contrary, confidently ahead in 1600x1200, providing an average frame rate of about 145fps whereas the Radeon X1900 XT CrossFire yields 130fps. The CrossFire system probably has worse performance scalability than Nvidia’s SLI here.
The game engine uses OpenGL and stencil shadows, so the GeForce 7900 GTX can enjoy a solid advantage of 20% over the Radeon X1900 XTX across all the resolutions. Two such graphics cards can get you as high as 120+ fps in 1600x1200, a feat that the Radeon X1900 XT CrossFire cannot repeat.
As for extreme FSAA modes, the GeForce 7900 GTX SLI allows playing with turned-on 8x SLI AA in any resolution, but the higher-quality 16x SLI AA is too difficult even for this platform as it yields a comfortable frame rate in 1024x768 only. The Radeon X1900 XT CrossFire seems much faster than the GeForce 7900 GTX SLI in that case, but you should be aware that 14x Super AA theoretically produces a lower-quality picture than 16x SLI AA with its honest super-sampling.
The gap between the GeForce 7900 GTX and the GeForce 7800 GTX 512 is getting smaller as the display resolution grows up because of the limiting effect of the graphics memory; the newer card has a slightly lower memory clock rate than the older one.
As for the multi-GPU configurations, the Radeon X1900 XT CrossFire is better in Call of Duty 2 thanks to its 96 pixel processors. The ring-bus memory controller gives the Radeon X1900 CrossFire an edge in the more resource-consuming 8x FSAA + 16x AF mode. This configuration makes the resolution of 1280x1024 playable, while the GeForce 7900 GTX SLI only allows playing the game with comfort in 1024x768. The maximum level of antialiasing is hardly available here since the two GeForce 7900 GTX deliver an average frame rate of less than 50fps in 1024x768.
Curiously enough, the performance of the single GeForce 7900 GTX almost equals the performance of the single GeForce 7800 GTX 512. The big difference in their GPU clock rates doesn’t seem to matter anything here. The memory frequency is not a bottleneck, either, or the GeForce 7900 GTX would be slower than the GeForce 7800 GTX 512 in high resolutions. Some peculiarity of the GeForce 7800/7900 architecture we are not aware of must have led to such results.
Overall, the GeForce 7900 GTX outperforms the Radeon X1900 XTX, and the GeForce 7900 GTX SLI is faster than the Radeon X1900 XT CrossFire, except in the 14x/16x FSAA + 16x AF mode where the memory load is too high even for the GeForce 7900 GTX SLI. Anyway, this mode is playable in 1024x768.
Oblivion is the continuation of the famous Bethesda Softworks «The Elder Scrolls» role series that includes such games as Arena, Daggerfall and Morrowind. I would say that The Elder Scrolls IV deserves to be called one of the most impatiently awaited projects of the 2006. The game can boast stunning graphics, which is not absolutely impeccable yet: there are a few bugs there. Namely, the level of landscape detail drops down significantly as the player moves away from the objects, although everything looks almost ideal at a shorter distance thanks to the newest 3D graphics engine that uses long pixel shaders but doesn’t involve Shader Model 3.0. This game also owes its great level of detail to powerful middleware generating faces and vegetation.
TES IV: Oblivion is based on a significantly modified and improved Gamebryo 3D engine developed by Emergent Game Technologies. The realistic vegetation is created by SpeedTree engine developed by Interactive Data Visualization. And the faces of all characters are created by a special FaceGen engine. The game is full of complex shader effects and HDR and together with high level of detail and scalability of the Oblivion world it makes the game extremely demanding for the graphics subsystem resources.
We decided not to test our solutions with enabled FSAA, because the HDR support gets disabled even on ATI cards in this case and the graphics quality drops down significantly. Also when HDR is on, Bloom effect cannot be enabled, and vice versa. The current version of ATI Catalyst drivers doesn’t have CrossFire support for TES IV. To enable this feature you have to rename the Oblivion.exe file into AFR-FriendlyD3D.exe, otherwise, no performance boost will occur. Since the game doesn’t have any built-in testing tools, we had to resort to FRAPS. To get a more detailed performance report we tested each card twice: in the open and inside the Imperial City jail tunnel.
In low resolutions you can see that GeForce 7900 GTX is slightly faster than Radeon X1900 XTX, although later on this advantage disappears because of the more efficient ring bus memory controller of the latter. Despite the enormous amount of shaders in the game, Radeon X1900 does not really benefit from 48 pixel processors it has. Unfortunately, we do not know all the peculiarities of the Oblivion graphics engine that is why we cannot tell you with all certainty what slows down the Radeon X1900 XTX: fewer TMUs than by GeForce 7900 GTX or some other factors.
As for the gaming performance, none of the single-card configurations can guarantee 60fps performance rate. In the best case the average performance level is around 45fps, but sometimes the performance can drop below 20fps. Note: Radeon X1900 XTX provides a much higher minimum performance rate than GeForce 7900 GTX. Multi-GPU support enlarges the list of supported resolutions. Moreover, in this case we see the influence of the CPU, even though our system is running on an Athlon FX-60. Even this powerful dual-core processor cannot cope with the open spaces of the Oblivion world.
The dependence on the processor performance can also be observed on the tunnel levels, although here it is not that huge: the performance level here is much higher and rests around 75fps. Other than that, the picture is very similar to what we have already seen: there is a clear parity between the new Nvidia solution and ATI Radeon X1900 XTX. In multi-GPU systems this parity is broken in 1600x1200 where two GeForce 7900 GTX cards cannot reach the performance bar even though they are only 10% behind the Radeon X1900 XT CrossFire system.
So, we have to admit that despite very high graphics subsystem requirements the performance in TES IV: Oblivion can sometimes still be limited by the CPU speed. Even ATI Radeon X1900 XT CrossFire or Nvidia GeForce 7900 GTX SLI may not help you out here.
There’s not much difference between the GeForce 7900 GTX and the GeForce 7800 GTX 512 in Far Cry , either. The SLI configuration with two new graphics cards from Nvidia is successfully competing with the Radeon X1900 XT CrossFire, except for the SLI AA/Super AA modes. The GeForce 7900 GTX SLI does yield a minimum frame rate of 60fps even with turned-on 8x SLI AA.
The new best-performance graphics card from Nvidia isn’t actually much better than the older one on the Research map as well, although this level is less CPU-dependent and can put the graphics card’s pixel processors to good use. The Radeon X1900 XTX is still in the lead among single-GPU solutions, outperforming the GeForce 7900 GTX by 10%.
The test scene goes on indoors most of the time, so the average performance of the cards is higher than in the previous case. The GeForce 7900 GTX SLI provides a comfortable speed in 1280x1024 with enabled 16x SLI AA, for example. Remember, however, that as soon as you go out into the open, the speed of the game may plummet to below comfortable level.
The GeForce 7900 GTX has an excellent result in Far Cry’s HDR mode, surpassing the Radeon X1900 XT CrossFire in high resolutions. The GeForce 7800 GTX 512 isn’t far behind, though, despite the lower graphics core clock rate – the newer card from Nvidia is only 3-4fps faster at best.
Just to remind you, the high dynamic range mode of Far Cry still doesn’t work well on graphics cards with ATI Technologies’ GPUs – their performance is far from optimal then.
The GeForce 7900 GTX performs in the same way on the Research map. You can play the game in 1600x1200 using HDR.
Unfortunately, you cannot improve the picture quality further by turning on full-screen antialiasing. The GeForce 7 architecture cannot smooth textures in FP16 format and thus doesn’t allow using FSAA and HDR at the same time. This option is available with ATI’s Radeon X1000 series, but ATI’s cards cannot filter such textures and have to emulate filtering by means of a special shader.
A shaders-saturated game, F.E.A.R. is sensitive to any increase in the pixel processors performance which directly depends on the GPU clock rate. This explains why the GeForce 7900 GTX is so much faster than the GeForce 7800 GTX 512 here. The new graphics card is as fast as two GeForce 7800 GTX working in SLI mode! The Radeon X1900 XTX is much lower than the GeForce 7900 GTX in low resolutions, but a little faster in high ones thanks to the more advanced memory controller. It should be noted that the performance of GeForce cards has grown considerably in F.E.A.R. with the driver suite released for the GeForce 7900.
We get a different picture when we compare the GeForce 7900 GTX SLI and the Radeon X1900 XT CrossFire: Nvidia’s solution does better in higher resolutions, at least in the 4x FSAA + 16x AF mode. The graphics memory performance has the biggest effect on the results at higher levels of antialiasing, and the GeForce 7900 GTX SLI is equal to the Radeon X1900 XT CrossFire in the 8x FSAA + 16x AF mode beginning from 1280x1024 resolution. Strangely enough, the SLI configuration with two GeForce 7900 GTX successfully contends with the Radeon X1900 XT CrossFire in two out of three resolutions in the 16x SLI AA mode, despite the use of super-sampling. Alas, neither of the multi-GPU configurations delivers a playable frame rate in this mode.
The speed of all the graphics cards is limited by the CPU performance in low resolutions, even though our testbed features an Athlon 64 FX-60. But we can see in 1600x1200 that the GeForce 7900 GTX is no better than the GeForce 7800 GTX 512. Each of the two cards delivers an average speed of 90fps, the Radeon X1900 XTX being 12-13% ahead of them both.
The GeForce 7900 GTX SLI configuration behaves somewhat oddly in the SLI AA modes. It should have obtained an advantage over the Radeon X1900 XT CrossFire by having more TMUs, but it has none. On the contrary, there’s a gap between these two solutions up to 50% in the 8x FSAA + 16x AF mode, and there’s a small performance gain over the GeForce 7800 GTX SLI, no bigger than 10%-30%.
As for the 14x/16x FSAA + 16x AF mode, the Radeon X1900 XT CrossFire is beyond competition in it, while the GeForce 7900 GTX SLI allows using resolutions no higher than 1280x1024.
Unlike the original Half-Life , the Lost Coast demo features complex special effects (like HDR emulation) created by means of pixel shaders. The GeForce 7900 GTX shows all its might here and leaves the GeForce 7800 GTX 512 behind by 10-20%. The SLI configuration with two GeForce 7900 GTX is unrivalled, too, as it delivers over 70fps where the Radeon X1900 XT CrossFire cannot yield a playable frame rate.
The GeForce 7900 GTX SLI platform is also in the lead in the 8x FSAA + 16x AF mode, but 1280x1024 resolution is the only one that seems to be playable. A comfortable frame rate is only achieved in the 14x/16x FSAA mode in 1280x1024 resolution, and the GeForce 7900 GTX SLI is slower than the Radeon X1900 XT CrossFire then.
Multi-GPU technologies from ATI and Nvidia worked incorrectly in this game, at least with the current drivers. Our turning SLI or CrossFire mode on led to a considerable performance hit.
The game engine is rather simple, so the GeForce 7900 GTX cannot profit much by its 100MHz higher core clock rate. Still, it is a little faster than the GeForce 7800 GTX 512.
The GeForce 7900 GTX SLI is not any worse than the Radeon X1900 XT CrossFire in the 8x FSAA + 16x AF mode, but cannot help falling behind it in the 16x SLI AA mode due to super-sampling. The average performance of the GeForce 7900 GTX SLI is about 70fps in the most popular resolution of 1280x1024, which is quite enough for comfortable play.
We can’t get a clear picture of how the graphics cards compare with each other in low resolutions. Even though we use antialiasing, modern graphics cards of the highest class are obviously limited by the CPU in this game. But we can see that the GeForce 7900 GTX is 8-9% faster than the GeForce 7800 GTX 512 in 1600x1200.
When there’s higher-level antialiasing to be done, the GeForce 7900 GTX SLI shows its best and is only inferior to the Radeon X1900 XT CrossFire in 1280x1024 and higher resolutions of the 16x SLI AA mode. Well even in that case the speed of the GeForce 7900 GTX SLI isn’t lower than 61fps, so you can play the game comfortably using the highest-quality full-screen antialiasing available in SLI mode.
Serious Sam 2 is full of shaders with multiple texture lookups, so the higher scene fill rate of the GeForce 7900 GTX comes in handy here, especially in lower resolutions. The lower memory frequency of the new graphics card hinders it in higher resolutions and drags it down closer to the level of the GeForce 7800 GTX 512. Well anyway, the GeForce 7900 GTX has the best results among all the participating single graphics cards.
Two such graphics cards in SLI mode give you a comfortable frame rate in all resolutions with 4x FSAA and in 1024x768 with 8x SLI AA. As usual, the Radeon X1900 XT CrossFire is suddenly ahead as soon as we switch to the 14x/16x FSAA mode.
Although there are a lot of pixel shaders in Splinter Cell: Chaos Theory , the GeForce 7900 GTX is hardly any faster than the GeForce 7800 GTX 512 and they are both behind the Radeon X1900 XTX beginning from 1280x1024 resolution. There must be something in the architecture of the G70 and G71 chips to produce such results; the performance of the GeForce 7900 GTX is somewhat limited by its lower memory frequency, too.
The GeForce 7900 GTX SLI offers a smaller speed reserve than the Radeon X1900 XT CrossFire platform does due to an obvious reason: the latter has 96 pixel processors in total against the Nvidia solution’s 48. The SLI platform offers a barely playable frame rate in the 8x SLI AA mode, while the Radeon X1900 XT CrossFire offers more than enough speed then.
The Pacific Fighters engine was originally developed with the GeForce 6/7 architecture in mind. For example, the game requires vertex texturing for the maximum quality of the water surface. Graphics cards on ATI Technologies’ GPUs have always been slower in this game than Nvidia’s, and this test session doesn’t contradict our earlier experience.
Having a higher scene fill rate, the GeForce 7900 GTX leaves the GeForce 7800 GTX 512 far behind and makes the resolution of 1600x1200 available for comfortable play with such conveniences as 4x FSAA and 16x anisotropic filtering. Owners of two GeForce 7900 GTX cards will even be able to use 16x SLI AA: an average speed of 67fps with a minimum of 38fps is quite enough.
This game can work with enabled Super AA or SLI AA, but it doesn’t make a good testing tool since its speed varies but too wildly. We don’t want to mislead you and do not publish the test results in these modes.
The GeForce 7900 GTX can show itself to advantage here. The game uses rather complex SM3.0 effects, so the higher GPU clock rate of the new graphics card puts it ahead of the GeForce 7800 GTX 512 and very close to the Radeon X1900 XT in 1600x1200.
Multi-GPU technologies do not provide any speed bonuses in Age of Empires 3 , but even worsen the speed as CrossFire does in 1280x1024.
Graphics cards of the GeForce 7 family have always been victorious in Dawn of War as they can efficiently process large textures and feature UltraShadow II technology which greatly accelerates stencil shadows rendering. The GeForce 7900 GTX feels well under such conditions, being only second to the Radeon X1900 XT CrossFire in 1600x1200 where its memory, slower than that of the GeForce 7800 GTX 512, begins to stifle its speed.
The GeForce 7900 GTX SLI subsystem put on a glorious performance, outpacing the Radeon X1900 XT CrossFire in every mode and resolution. The only exception is the resolution of 1600x1200 with turned-on 16x SLI AA: the memory load is too high here for the GeForce 7900 GTX SLI to keep on being superior to the Radeon X1900 XT CrossFire with its somewhat easier 14x Super AA mode. The performance of both the multi-GPU platforms is far from optimal and you don’t get a comfortable frame rate with either of them at the highest level of antialiasing.
This test is rather remote from today’s gaming realities and its results are only interesting from a purely theoretical point of view. We can note once more that the GeForce 7 architecture is superbly adapted for applications that do not require a high pixel shader performance. The GeForce 7900 GTX is faster than the Radeon X1900 XTX in all resolutions with turned-on 4x FSAA, even though the gap isn’t bigger than 8-10%.
As for the harder antialiasing modes, those that you may only venture to enable on a dual-chip configuration, the GeForce 7900 GTX SLI is losing its ground steadily in high resolutions of the 8x FSAA mode and then in the 14x/16x FSAA mode where there’s a 60% gap between the two graphics subsystems. The GeForce 7900 is not to be blamed for this since 16x SLI AA is much harder on the graphics memory than 14x Super AA.
The GeForce 7900 GTX SLI stops a mere 412 points short of the Radeon X1900 XT CrossFire whereas the GeForce 7900 GTX is as many as 781 points behind the Radeon X1900 XTX. So, there’s no triumph for Nvidia in 3DMark05, but the company’s new solution cannot be considered as utterly defeated, either. Let’s better view the results of the separate tests.
In the first test the GeForce 7900 GTX is about 10-12% slower than the Radeon X1900 XTX using 4x FSAA. The two multi-GPU configurations, on ATI Radeon X1900 XT CrossFire and Nvidia GeForce 7900 GTX SLI, have the same performance in this mode, hitting against the CPU-imposed speed ceiling.
The new SLI subsystem from Nvidia is slower than ATI’s CrossFire platform in the higher FSAA modes due to the differences in their memory controllers. The gap isn’t bigger than 5-10% in the 8x FSAA mode, though.
The top-end graphics cards have the same speeds in the second test. The GeForce 7900 GTX isn’t much worse than the Radeon X1900 XT; the gap isn’t bigger than 10% at worst.
The multi-GPU tandems behave much alike to themselves in the first test, but the GeForce 7900 GTX SLI looks somewhat more confident in the 8x FSAA + 16x AF mode when earlier. It even overtakes the opponent in 1024x768 and is no farther than 9-10% behind it in 1600x1200. It’s logical to assume that such results are due to the small size of the scene that puts a smaller load on the memory subsystem.
In the third test, on the contrary, the Radeon X1900 XTX and the Radeon X1900 XT CrossFire claim a victory among single-card and multi-GPU solutions, respectively. Note that it’s in the third test that the difference between the GeForce 7900 GTX and the GeForce 7800 GTX 512 is the smallest, so once again the unknown factor we have referred to above is at work again, limiting the performance of the new graphics card. Moreover, the third test scene is large, so its speed depends more on the graphics memory performance than in the two previous tests.
The results of the new graphics card from Nvidia are confirmed in each separate test even with enabled full-screen antialiasing. The GeForce 7900 GTX is nowhere in the lead, but never falls too far behind the ATI Radeon X1900 XTX, either.
The Radeon X1900 XTX has to leave the top place in 3DMark06 to Nvidia’s new graphics card which has scored 142 points more. Not much of a difference, but the GeForce 7900 GTX is the leader nevertheless. Note also that the performance has grown much relative to the GeForce 7800 GTX 512 – 580 points is quite a lot for 3DMark06. This is expectable since 3DMark06 makes wide use of shader-based visual effects.
There’s a smaller difference between the multi-GPU configurations on Radeon X1900 XT and GeForce 7900 GTX than between the respective single cards and amounts to 31 points only. So, we should regard these two top-end graphics solutions from Nvidia and ATI as equivalent in 3DMark06.
The Radeon X1900 architecture cannot show its very best in the SM2.0 tests because they do not use version 3.0 shaders. Moreover, the first test is a very large scene and seems to be the reason why the GeForce 7900 GTX has a higher overall score than the Radeon X1900 XTX. We’ll check this supposition shortly.
The GeForce 7900 GTX finds it more difficult to crunch through the SM3.0/HDR tests. It is slower than the Radeon X1900 XTX with its 48 pixel processors and Fetch4 technology.
The GeForce 7900 graphics processor cannot do full-screen antialiasing and work with HDR all the same time, so we publish the SM2.0 tests results for the 4x FSAA + 16x AF mode only. The performance is too low at higher levels of antialiasing, which is no good for the measurement accuracy.
As we had suspected, the GeForce 7900 GTX came out on top in the first SM2.0 test, leaving the Radeon X1900 XTX behind in two out of three display resolutions. The new card just wanted a little more of memory subsystem performance to win the resolution of 1600x1200, too. Although the memory frequency of the Radeon X1900 XTX is somewhat lower, it is governed by a more efficiency ring-bus controller whereas the GeForce 7900 GTX uses a traditional controller.
We see the same as we compare the Radeon X1900 XT CrossFire with the GeForce 7900 GTX SLI. The former overtakes the Nvidia platform by a narrow margin in 1600x1200 resolution only. In the lower resolutions, the ATI platform is about 10% slower.
The second SM2.0 test is less sensitive to the texturing speed than the first one, so there’s not much use from the 24 TMUs here. This is why the GeForce 7900 GTX is only equal to the Radeon X1900 XTX in 1024x768 and falls behind in higher resolutions, by up to 15%. The same goes for the struggle between the two multi-GPU platforms, too.
So, our supposition about the overall scores of the SM2.0 tests has been proved by the results of the separate tests. The GeForce 7900 GTX really owes much of its overall advantage to the first, large-scale test.
Now that we’ve tested the Nvidia G71 (GeForce 7900) graphics processor and the GeForce 7900 GTX graphics card who are we to call the king of 3D? Is the new graphics card from Nvidia better than the Radeon X1900 XTX?
The new chip is very good beyond doubt. Thanks to the new, thinner tech process and the optimized design of the pixel processors Nvidia has come up with a top-end graphics chip which is the smallest and most economical in the industry. Although the G71 consists of only 278 million transistors – 24 million less than in the G70 – it is quite capable of rivaling the much more complex ATI R580 (Radeon X1900). The new GPU doesn’t bring about anything particularly new at that; it is in fact a GeForce 7800 improved and working at a higher clock rate. The GeForce 7900 differs from its predecessor in three things only: it consists of fewer transistors, it is manufactured on a thinner tech process, and it supports dual-link DVI.
The chip embodies Nvidia’s approach to making GPUs, which is to keep the number of pixel processors and TMUs in balance. The G71 has 24 of either, while the Radeon X1900 has 48 pixel processors and only 16 TMUs. Our tests have shown that Nvidia’s approach is justifiable even in modern games with their abundance of pixel shader-based visual effects because the Radeon X1900 often slows down under high textural load. You should also keep it in mind that though the Radeon X1900 XTX may be architecturally better suited for future games, its raw performance may prove too low to run such games at an acceptable speed when they do come out.
So, we can’t say the GeForce 7900 GTX is a clear winner today, but it has done at least no worse than the Radeon X1900 XTX in 14 out of 18 tests. And it has the advantage of lower price: $499 against $549. ATI’s graphics card, however, does better in high resolutions as well as in FSAA modes higher than 4x thanks to its more advanced memory controller. It is not quite clear why the memory frequency of the GeForce 7900 GTX has been set lower than that of the GeForce GTX 512, especially as the new graphics card carries memory chips capable of working at frequencies up to 900 (1800) MHz. This must have been made to ensure a higher stability of the device, but its performance has obviously suffered: there’d surely be a higher performance growth relative to the GeForce 7800 GTX 512 if it were not for the reduced memory clock rate.
Talking about the GeForce 7900 GTX it is impossible to omit its very low power consumption (for a top-end graphics card), which is about 85 watts under load. A SLI configuration with two such cards is going to consume 170W as opposed to 240W consumed by a Radeon X1900 XT CrossFire system. Particularly it means that a lower-wattage and less expensive power supply will be required for a computer with a GeForce 7900 GTX SLI graphics subsystem than for a Radeon X1900 XT CrossFire.
So, the new flagship product from Nvidia looks competitive enough, even though it has not become an indisputable performance leader. It’s now up to Nvidia to satisfy the demand for this highly appealing graphics card which is going to be high, considering the price tag. We hope the GeForce 7900 GTX will not repeat the fate of the GeForce 7800 GTX 512 and will come to market in mass quantities very soon.
Highs:
Lows: