by Alexey Stepin , Anton Shilov, Yaroslav Lyssenko
12/13/2006 | 10:03 AM
November 8, Nvidia announced its new graphics processor G80 along with two G80-based graphics cards, a flagship GeForce 8800 GTX and a less advanced, yet more affordable GeForce 8800 GTS. They have come to replace the GeForce 7900 GTX and the dual-chip GeForce 7950 GX2 models, the last-generation solutions based on an architecture that traces its origin back to the 2-year-old NV40 processor and even incorporates some features of the still older NV35.
<%BANNER[article]%>We discussed the G80 architecture in this review in which we scrutinized GeForce 8800 GTX, the senior model of the family. We also covered the problems of image quality and performance in theoretical benchmarks. The new graphics card did well in every case, surpassing last-generation solutions such as ATI Radeon X1950 XTX and Nvidia GeForce 7900 GTX.
So, there was no doubt about the huge potential of the new solution from Nvidia, yet we couldn’t make our final verdict because the theoretical benchmarks didn’t show the full picture. In such benchmarks the whole computing power of the unified-architecture GPU is dedicated to solving only one problem, which is not the case with real applications, i.e. with 3D games, where performance of any graphics solution is determined by a number of various factors such as the specifics of the particular game engine and all the technical characteristics of the GPU such as its memory bandwidth and memory controller design, the number of TMUs and ROPs, core clock rate and shader processor design.
What we mean to say is that there is no quick and simple way to find out how good a new graphics card is going to be under real gaming conditions. You can only check this out by testing it in different games of different genres. That’s what we’ve done with our GeForce 8800 GTX.
Since we had to deal with a completely new architecture as well as with a graphics card that claimed to be the fastest solution on the market, we decided to benchmark it in more games than usual to give you a more detailed picture of its performance.
We want to apologize for a month-long delay in preparing this important review. It is due to some troubles we encountered when testing the graphics card. We’ll tell you about the nature of the troubles at the end of the article. Right now, let’s see what the GeForce 8800 GTX can do!
We used the following hardware for this test session:
The graphics card drivers were set up in such a way as to provide the highest possible quality of texture filtering.
ATI Catalyst:
Nvidia ForceWare:
We selected the highest possible graphics quality level in each game using only those settings that were accessible to the player. We didn’t modify the games’ configuration files that might contain additional settings. Performance was measured with the games’ own tools or, if not available, manually with Fraps 2.8.1. When measuring the speed of a card with Fraps, we chose simple routes in the game to achieve repeatability of results and reduce the measurement error. For the same purpose we would take a long enough fragment when using a cut-scene on the game engine. We also measured the minimum speed of the cards where possible because this is often the criterion that determines how comfortable it is to play a game.
We tested the cards in three standard resolutions: 1280x1024, 1600x1200 and 1920x1200. We didn’t use the “pure speed” mode (without full-screen antialiasing) because it is not relevant for top-end graphics cards. Instead, we used two additional high-quality antialiasing modes. The test session including both single cards and multi-GPU configurations, different antialiasing methods were used in the FSAA 6x/8x + Aniso 16x and FSAA 14x/16x + Aniso 16x test modes. They are listed in the table:
As you see, we focus on quality and choose the best-quality, although resource-consuming, 8x MSAA (called 8xQ in the Control Panel) and 16xQ CSAA modes on the GeForce 8800 GTX. The latter mode features a double number of color/depth samples in comparison with 16x CSAA
We enabled anisotropic filtering and standard FSAA modes from the games’ own menus. If not possible, we forced the necessary mode from the driver. The high-quality antialiasing modes were forced from the driver. On the GeForce 8800 GTX, if the game supported standard FSAA modes, we used the Enhance the application setting option described in our theoretical review. We benchmarked the cards without FSAA in those games that do not support it due to the specifics of their engines.
Besides the GeForce 8800 GTX, we included the following graphics cards into this review:
Multi-GPU systems were represented by the following configurations:
So, we’ve got the fastest single graphics cards as well as all modern dual- and even four-processor configurations for a more accurate comparison. Unfortunately, we couldn’t benchmark a SLI tandem made out of two GeForce 8800 GTX due to problems with our nForce4 SLI X16-based mainboard, but we are going to do that in an upcoming review.
We used the following games and benchmarks:
First-Person 3D Shooters
Third-Person 3D Shooters:
RPG
Simulators:
Strategies:
Synthetic Benchmarks:
This game still does not support non-standard antialiasing modes, so we benchmarked the cards at 4x FSAA + 16x AF only.
The GeForce 8800 GTX takes a good start, outperforming the Radeon X1950 XTX CrossFire system (by 10-12%), not to mention single last-generation cards.
The results of the GeForce 7950 Quad SLI platform are disappointing. Nothing has changed since our review of that technology – it is still a little slower than the single GeForce 7950 GX2 in this game.
None of the cards has any problems with performance here. You get no less than 66fps at 1600x1200 with enabled 4x FSAA even on a single GeForce 7900 GTX.
This sequel to the Battlefield series brings the player into the future, into the year of 2142, to be exact. And you find yourself right in the heat of a battle between two superpowers – the European Union and the Pan Asian Coalition. The game offers the player a number of exciting ways to pass away the time like piloting various battle machinery, including gigantic robots, or engaging into a team combat in the Titan mode. In this mode, the goal of a team of players is to protect a huge flying platform called Titan while attacking a similar enemy platform.
From a technical point of view, the game isn’t much different from the original as it runs on the same engine with certain modifications and only boasts more of special effects. Non-standard antialiasing modes are still not supported, so we had to limit our test to 4x FSAA.
The results are a good illustration of the potential of the GeForce 8800 GTX relative to its opponents.
It is not the GeForce 8800 GTX and even not the quad-processor GeForce 7950 Quad SLI, but the ordinary GeForce 7900 GTX SLI tandem that is the leader at the resolution of 1280x1024 pixels. At 1600x1200, however, it has to give way to Nvidia’s new graphics card that has a faster memory subsystem. Note that the GeForce 8800 GTX doesn’t lose much speed switching to the higher resolution whereas the speed of the 7900 GTX SLI in 1280x1024 and 1600x1200 differs dramatically.
The GeForce 7900 GTX is the only single graphics card to deliver too low performance in 1600x1200. The Radeon X1950 XTX and the GeForce 7950 GX2 both ensure a comfortable frame rate in this mode.
Besides the Radeon X1950 XTX CrossFire and GeForce 7900 GTX SLI configurations, the GeForce 8800 GTX is the only graphics solution capable of providing a comfortable frame rate in 1920x1200. The GeForce 7950 Quad SLI system cannot deliver it despite the loudly touted advantages its four GPUs are expected to give.
It is all quite different in 8x FSAA + 16x Aniso mode: notwithstanding its true 8x MSAA, the GeForce 8800 GTX enjoys a big lead over the rest of the participating graphics cards. The GeForce 7950 Quad SLI at last shows its worth, but its performance is barely enough for comfortable play even in 1600x1200 whereas the GeForce 8800 GTX provides a near-comfortable frame rate in 1920x1200.
The GeForce 8800 GTX is dominating in the highest-quality and hardest antialiasing mode, being the single graphics card to yield acceptable performance here. Its speed is high, although not reaching 60fps, even at 1920x1200.
We should note that CrossFire technology doesn’t work correctly in this game if you enable Super AA.
There’s no sense in comparing the numbers we’ve got in this test because all the graphics cards, except for the GeForce 7900 GTX, have roughly the same speed. The GeForce 8800 GTX is obviously unable to reveal its potential, being limited by the capabilities of the CPU and platform.
This test mode is harder, yet it makes no big difference. Note that the Radeon X1950 XTX with its true 6x multisampling is almost as fast as the GeForce 8800 GTX with 8x MSAA. The GeForce 7950 GX2 performs well singly and in pair (we mean the GeForce 7950 Quad SLI system based on two such graphics cards), but the 8xS mode of the single GeForce 7900 GTX is again proved useless for a real-life application.
The GeForce 8800 GTX is somewhat slower than the Radeon X1950 XTX CrossFire, providing a comparable quality of antialiasing, but we shouldn’t forget that the new graphics card loses to the fastest multi-GPU platform built on last-generation cards. Anyway, either solution allows using 16xQ CSAA and 14x Super AA whereas the GeForce 7950 Quad SLI can hardly meet our performance requirements even in 1600x1200, although Far Cry is by far not the hardest 3D application today.
What we’ve said above is true for both Pier and Research maps. The single difference is that the GeForce 8800 GTX looks somewhat more confident in the latter case, outperforming the Radeon X1950 XTX CrossFire greatly in 8x FSAA mode and losing less in 14x/16x FSAA mode.
The 384-bit bus influences the results strongly when we enable the game’s FP HDR mode. The CPU has a smaller effect on performance of graphics cards as the resolution grows up, and the GeForce 8800 GTX increases its lead over the Radeon X1950 XTX CrossFire, up to 30% in 1920x1200. The Pier map is heavy for the platform at large, that’s why the GeForce 8800 GTX doesn’t slow down much as the resolution grows up.
The Research map is less CPU-dependent than Pier, and the GeForce 8800 GTX enjoys a great advantage over the Radeon X1950 XTX CrossFire. Well, ATI’s solution suits for playing this game in every resolution too, including 1920x1200. The same is also true for the GeForce 7900 GTX SLI.
The GeForce 8800 GTX has the best result in two out of three resolutions, but it finds itself a little behind the GeForce 7950 Quad SLI in 1600x1200. And it is only 3fps ahead of the Radeon X1950 XTX in 1920x1200.
The GeForce 7950 Quad SLI system breaks away from the others and is the only graphics solution to provide over 60fps in 1920x1200. The Radeon X1950 XTX CrossFire acts up here: it has better results in 1920x1200 than in 1280x1024. It looks like this system doesn’t enable Super AA mode in the highest resolution, but uses 4x MSAA instead.
The GeForce 8800 GTX gains its ground in this test, but its performance is too low for comfortable play. There’s the same oddity with the Radeon X1950 XTX CrossFire here. The system now seems to be using 6x MSAA instead of 4x MSAA. This must be some problem in the ATI Catalyst driver and its support of multi-GPU technology in F.E.A.R.
We described this sequel to the popular 3D shooter F.E.A.R. in our earlier review. It runs on the same engine and has the same preferences as the original, but features much more advanced visual effects that increase the requirements to the graphics subsystem. Unfortunately, the game is not free from the drawback of earlier versions of F.E.A.R. as it does not support resolutions above 1600x1200.
In spite of the increased graphical complexity, the average performance of every card is somewhat higher here than in the original game. The GeForce 8800 GTX is an unrivalled leader in every display resolution supported by the game.
The speed is lower with enabled 8x MSAA, yet the minimum performance of the GeForce 8800 GTX is never below 25fps even in 1600x1200. It is only the GeForce 7950 Quad SLI that turns in a better result. The F.E.A.R. series is one of the few places where Nvidia’s quad-processor technology is indeed highly efficient.
It is not advisable to enable 16xQ CSAA here due to low performance. As you’ve seen above, there is a much higher speed with 8x MSAA while the antialiasing quality is very close to that provided by 16xQ CSAA.
The deferred rendering technique employed in this game makes it incompatible with full-screen antialiasing. That’s why there are only anisotropic filtering results here.
This game is so demanding that a comfortable frame rate in high resolutions could only be achieved on the senior version of the CrossFire platform. Now you can solve this problem with a single GeForce 8800 GTX. Take note of the min speed numbers, too – Nvidia’s new solution provides a much bigger performance reserve. Its min speed is as high as 50fps even in 1920x1200 which means smooth gameplay even in hardest scenes.
The GeForce 8800 GTX and the Radeon X1950 XTX CrossFire reach the highest speed possible for the CPU model our testbed is equipped with in low resolutions, and their speeds coincide in 1920x1200, too. Obviously, the 4x FSAA test mode cannot load such powerful graphics solutions fully here.
The cards differ more in 8x FSAA + 16x Aniso mode, starting from the resolution of 1600x1200 pixels. The GeForce 8800 GTX is a leader, followed by the GeForce 7950 Quad SLI. The Radeon X1950 XTX CrossFire and the GeForce 7900 GTX SLI are in the top four, too. Among single graphics cards with an ordinary, non-unified architecture, we can single out the Radeon X1950 XTX that provides 60fps in 1920x1200.
In the most resource-consuming, yet also the best-quality mode, the GeForce 8800 GTX is somewhat slower than the Radeon X1950 XTX CrossFire. The new card from Nvidia cannot give you 60fps in 1920x1200 as the more expensive solution from ATI does.
Strangely enough, it is in an OpenGL application where Nvidia’s cards have always used to prevail that the GeForce 8800 GTX doesn’t show anything exceptional. Yes, it is ahead of every other single card, including the dual-processor GeForce 7950 GX2, but is slower than the GeForce 7950 Quad SLI and even than the Radeon X1950 XTX CrossFire in high resolutions. It seems Nvidia’s programmers haven’t yet fully optimized the driver for the new graphics card series.
The GeForce 8800 GTX looks better here and is competing with the Radeon X1950 XTX CrossFire. Yet it is still slower than the GeForce 7950 Quad SLI and doesn’t give you a comfortable speed in resolutions above 1280x1024.
In spite of the impressive specs of the GeForce 8800 GTX, particularly of its extraordinary memory subsystem bandwidth, our turning on 16xQ CSAA mode leads to a performance hit below comfortable level. We guess that only SLI tandems made out of two GeForce 8800 GTX or GTS cards would provide an acceptable speed here.
Note the surprisingly high result of the Radeon X1950 XTX CrossFire system in 1920x1200 – it just doesn’t turn on 14x Super AA mode then.
It’s all similar to Prey in this test because both games run on the same engine, although in Prey it is modified to meet today’s requirements to graphics quality.
This game being less demanding than Prey , the GeForce 8800 GTX allows playing with comfort in 1600x1200 and even in 1920x1200, but the GeForce 7950 Quad SLI provides a bigger reserve of speed in the latter resolution.
As opposed to Prey , 16xQ CSAA is practicable here. You can even try to enable this mode in 1600x1200, although the speed will be just on the verge of comfortable. The GeForce 7950 Quad SLI is not that good. It only makes the resolution of 1280x1024 playable.
The problem with 14x Super AA on the Radeon X1950 XTX CrossFire shows up here, too.
In 4x FSAA + 16x Aniso mode the GeForce 8800 GTX outperforms the more expensive Radeon X1950 XTX CrossFire system only in 1280x1024, but its superiority over the single last-generation cards is impressive. However, the driver is evidently not yet optimized enough for the new architecture. The min speed of the GeForce 8800 GTX is very low, much lower than that of the single GeForce 7900 GTX.
It’s the same with the min frame rate, although the GeForce 8800 GTX is much faster than the others in terms of average speed. So, you can try to play the game in 1600x1200, but the slowdowns to 8-10fps won’t make your gaming experience very comfortable.
The same thing happens in 16xQ mode: the low minimum of speed of the GeForce 8800 GTX doesn’t depend on the FSAA mode you use.
So, it’s a kind of paradox here: providing a high average frame rate, the new flagship card from Nvidia is in fact unsuitable for playing Serious Sam 2 . We hope this will be corrected in the next ForceWare release.
The new-generation technology shows its very best here – the GeForce 8800 leaves the rest of the participating graphics cards far behind. In the highest resolution this card has a min speed of 44fps which is higher than the average frame rate of single-chip last-generation solutions and by 30% higher than the min speed of the GeForce 7950 Quad SLI. This guarantees you smooth gameplay without slowdowns and jerkiness.
Note also the low minimum speed of the GeForce 7950 Quad SLI in high resolutions and the lack of any speed growth with the ATI Radeon X1950 XTX CrossFire.
The Radeon X1950 XTX CrossFire works correctly in Super AA mode and has an average frame rate comparable to that of the GeForce 8800 GTX in high resolutions. We shouldn’t forget, however, that we are talking about a multi-GPU tandem that requires a special mainboard and costs much more than the new-generation graphics card from Nvidia.
It’s only in 1280x1024 that the GeForce 8800 GTX is faster than the Radeon X1950 XTX CrossFire. In the higher resolutions these two graphics solutions deliver comparable speeds, possibly limited by the speed of the TMUs and ROPs. The new card offers a higher minimum of speed, though.
The new card from Nvidia suddenly loses to the ex-flagship GeForce 7950 GX2. The game abounds in special effects that involve doing a lot of math and this should match the GeForce 8800 GTX with its huge computing power perfectly, so the only explanation we can think of is that its driver has not yet been thoroughly optimized.
It’s somewhat better for the GeForce 8800 GTX in 8x FSAA mode, yet its performance is still far from what we might expect. The card is slower than the single Radeon X1950 XTX in every resolution and the gap is growing wider in higher resolutions. We should make allowances for the GeForce 8800 GTX doing 8x MSAA, which is more resource-consuming than the Radeon X1950 XTX’s 6x MSAA, yet the defeat of the Nvidia card looks strange and surprising anyway.
The GeForce 8800 GTX can’t win in the most extreme antialiasing mode, either, but it is at least ahead of the GeForce 7950 Quad SLI and delivers comfortable performance in resolutions up to 1600x1200.
Tomb Raider: Legend produces a picture that is quite the opposite of the previous test. The GeForce 8800 GTX makes a brilliant display of its tremendous strength, humiliating its opponents and allowing to use 4x FSAA even in the resolution of 1920x1200 pixels. The Next Generation Content option of this game speaks for itself. Indeed, only new-generation GPUs can run the game in this max-quality mode at a sufficiently high speed.
The GeForce 8800 GTX even makes it possible to enable 8x MSAA in this game, except for extreme resolutions like 1920x1200 or 2560x1600.
Considering that third-person shooters are quite playable if you’ve got an average frame rate of 45-50fps, the new graphics card is capable of running this game at a comfortable speed even with enabled 16xQ CSAA. And you only need a single $599 card for that, not a multi-GPU tandem costing over $1000. Nvidia’s new flagship graphics card is indeed amazing!
Gothic 3 belongs to the same variety of role-play games as TES IV: Oblivion . You’ve got the main character under control and are looking at the world from out of his eyes. The game runs on the Genome Engine constructed specifically for Gothic 3 by Pluto 13, the developer of the game. We guess the game is inferior to Oblivion in its visuals, but tries to conceal this by making wide use of such effects as Bloom and Depth of field as well as of haze to cover up the imperfections. Still, the detailness of the game world is high enough in Gothic 3 to be considered comparable to TES IV .
Unfortunately, the game engine is yet imperfect. The latest version (1.09) as of the time of our writing this review didn’t support FSAA although the game didn’t use FP HDR. The imperfection is also indicated by the extremely high requirements of the game to the graphics card resources at max graphics quality settings. Just look at the numbers:
The last-generation graphics cards and multi-GPU systems based on them can hardly surpass a 30fps mark even in 1280x1024. The GeForce 8800 GTX with its 44fps looks like a breakthrough. Well, Gothic 3 seems to be quite playable subjectively even at a speed of 30fps.
All the multi-GPU systems, from the GeForce 7950 Quad SLI to the Radeon X1950 XTX CrossFire with its hardware frame compositing unit, work incorrectly, which is yet another indication of how imperfect the game engine is. There is flickering and loss of shadows, so the owner of a multi-GPU platform will have to disable SLI or CrossFire mode to play Gothic 3 normally.
This game continues the Neverwinter Nights series and belongs to a different variety of role-play games than Gothic 3 and TES IV: Oblivion . There is a whole gang of characters under the player’s control here. This subgenre of computer RPGs may be regarded as the classic one, tracing its origin back to the legendary Baldur’s Gate . The gameplay of NWN 2 deserves a full-size review unto itself, and as for technicalities, the game can boast a rather advanced graphics engine that supports normal maps and HDR.
The game doesn’t support FSAA. It is possible to force it using some tricks, but we do not tweak games’ engines, using only their standard settings instead. That’s why we benchmarked the cards in NWN 2 with enabled HDR but without full-screen antialiasing. We used one of the cut-scenes recorded on the game engine, long and rather heavy for the graphics subsystem.
The top-end graphics cards all have very low speeds at the maximum graphics quality settings, but the specifics of the game are such that the gameplay is smooth and comfortable even at 30fps. You can play even at a lower speed, but it must not be lower than 15fps. Otherwise, there may be errors in controlling the characters: they miss your orders, “forget” to strike blows, etc.
There are only two graphics cards to meet the mentioned requirement: GeForce 7950 GX2 and GeForce 8800 GTX. And the former card only makes the resolution of 1280x1024 playable. The other participating devices have humbler results (but this low performance will only be observed in scenes with abundant vegetation and with a lot of models).
At the time of our tests the game had a number of defects, besides its too high requirements to the graphics card. There are problems with support for multi-GPU technologies and there is incompatibility with Radeon X1000 cards (such effects as water reflections and water refractions do not work on them). We hope this will be corrected by the game developer with patches.
There seem to be nothing extraordinary about the results of the GeForce 8800 GTX, yet it provides the same performance as the more expensive Radeon X1950 XTX CrossFire subsystem. Let’s now also see how the new card will behave in open game scenes which are terribly heavy for any graphics solution.
The GeForce 8800 GTX provokes a small sensation here. It is the first time we see a single graphics card to deliver an average frame rate of near 60fps in 1920x1200! You cannot get that even on the Radeon X1950 XTX CrossFire tandem, not to mention weaker solutions. The min speed of the new card is high enough at about 30fps, so you are guaranteed a smooth gameplay.
Besides benchmarking the GeForce 8800 and its competitors in the high dynamic range (HDR) mode without FSAA (since the GeForce 6/7 do not support HDR with full-scene antialiasing), we also decided to see how the new architecture from Nvidia performs with FSAA and HDR enabled in Oblivion game.
With FSAA 4x and anisotropic filtering 16x activated, the GeForce 8800 GTX demonstrates exceptional performance in all resolutions, allowing gamers to enjoy both HDR and antialiasing.
In fact, in all three cases the model 8800 GTX performs inline with two Radeon X1950 XTX in CrossFire mode, which is a respectable level of performance. The single Radeon X1950 XTX does not deliver comfortable framerate in any resolution, which proves once again that the HDR + AA mode is a too hard task for single-chip previous-generation hardware.
If we take the GeForce 8800 GTX to the next level of eye candy, with FSAA 8x and AF 16x, we will witness that even such a powerful graphics accelerator cannot deliver 60fps in 1600x1200 and 1920x1200 resolutions. Nevertheless, performance in 1280x1024 is tangibly higher compared to the Radeon X1950 XTX CrossFire, not talking about the single Radeon X1950 XTX that cannot keep up with the GeForce 8800 GTX even rendering the frames with lower quality FSAA.
We do not know why, but in those resolutions speed difference between the three competing solutions is negligible, which means that there is something, perhaps, in the engine of the game itself, that limits the number of frames per second with MSAA 6x or 8x activated.
It is impressive that even with CSAA 16x in 1280x1024 resolution enabled the GeForce 8800 GTX still delivers good framerates. In higher resolutions, however, the speed is not high enough for comfort game play.
Performance of the Radeon X1950 XTX in tandem is pretty strange here: on the one hand, it is obvious that FSAA 14x is activated (due to the fact that performance in 1280x1024 is lower than that on the previous graph), but on the other hand, the speed does not decrease as resolution rises (there is a 6fps drop in 1600x1200 though, which we cannot explain).
Despite of the fact that test conditions are much harder in case of outdoor levels of Oblivion, the GeForce 8800 GTX still shows good scores in 1280x1024 and 1600x1200 resolutions and is generally ahead of competing single-chip or dual-chip Radeon X1950 XTX solutions. In fact, even in 1920x1200 the novelty can beat the rivaling X1950 XTX CrossFire while still providing more or less normal framerate.
It is interesting to note that the GeForce 8800 GTX can perform really well even with FSAA 8x and anisotropic filtering 16x activated in outdoor levels of Oblivion, a respectable achievement. The GeForce 8800 GTX, in fact, guarantees nearly 100% improvement of minimal fps when compared to the Radeon X1950 XTX!
Activation of 16x CSAA mode does not reduce performance of the GeForce 8800 GTX dramatically: framerates in 1280x1024 are still smooth and even performance in 1600x1200 should be satisfactory for non-hardcore gamers.
Such a slight drop in performance when shifting from MSAA 8x to CSAA 16x proves that the latter is a very well balanced combination of quality and processing demands.
When it comes to Super AA 14x and the Radeon X1950 XTX CrossFire, we again see no performance difference between 1280x1024 and 1920x1200 resolution as well as a slight performance drop in 1600x1200.
Quite expectably, the GeForce 8800 GTX shows its best in high resolutions thanks to its wide memory bus. It is faster than the ex-leader GeForce 7900 GTX SLI then.
The GeForce 8800 GTX passes the test successfully with enabled 8x MSAA, too. In 1920x1200 the card provides a frame rate that would be comfortable for any first-person 3D shooters, not to mention Titan Quest with its more moderate requirements.
The new graphics card doesn’t slow down much even in 16xQ CSAA mode, making it possible to play the game with the highest possible quality of antialiasing.
The ATI Radeon X1000 family doesn’t support vertex texturing and cannot render the water surface with the highest quality in this game. This is only possible on Nvidia’s graphics cards, including the new GeForce 8800 series.
The flaws in the driver that lead to the GeForce 8800 GTX having low performance in OpenGL games like Quake 4 or Prey affect its performance in Pacific Fighters , too. And still, the new card is no worse than Nvidia’s best last-generation solutions here.
The minimum of speed of the GeForce 8800 GTX goes lower here than that of the GeForce 7950 GX2, which confirms our supposition about some flaws in the driver.
It’s different in 16x FSAA mode: the last-generation graphics cards cannot cope with the increased load while the GeForce 8800 GTX stands its ground even in 1920x1200 thanks to its economical CSAA algorithm and wide memory bus.
There are some problems with the GeForce 8800 GTX in X3: Reunion , too, although this is not an OpenGL application. The new card has a high enough average speed, but occasionally slows down to 10-15fps. This is yet another indication that the driver for the new GeForce series still needs improvement. It is possible that this fluctuating performance is due to errors in the operation of the scheduler that is responsible for allotting the shader processors to execute pixel and vertex shaders.
We can see the same problem in 8x MSAA mode, although the min speed is somewhat higher here and the average speed is almost as high as that of the Radeon X1950 XTX CrossFire in high resolutions.
Notwithstanding the impressive potential the GeForce 8800 GTX has shown in many other games, it is the Radeon X1950 XTX CrossFire that suits best for playing X3 with the highest possible FSAA. It is due to those sudden fluctuations of performance from 44-55fps to 12-18fps. The speed of the ATI solution doesn’t vary so wildly and it ensures a more comfortable and smoother gameplay.
The War Chiefs is not a new game, but an add-on to the original Age of Empires 3 that includes new maps, campaigns and war units. You can now play for Indian civilizations such as the Aztecs, Iroquois, and Sioux, hence the name of the add-on. It’s beyond the scope of the review to describe the changes in game-play, but from the graphics side The War Chiefs is the same as the original and runs on the same engine, without any modifications. The game is still incompatible with FSAA modes higher than 4x.
Just as earlier, the GeForce 7950 Quad SLI system suffers a performance hit in comparison with the single card, which confirms that that there are no serious technical changes in The War Chiefs . As for the GeForce 8800 GTX, its average performance is on the same level with the Radeon X1950 XTX CrossFire, but its minimum speed is considerably higher than that of the ATI solution.
Company of Heroes is considered by many to be the best tactical RTS with a World War 2 setting, and not without reasons. Besides well-balanced gameplay, the game features superb visuals like in a regular first-person 3D shooter. Such a high level of detail can seldom be seen in a real-time strategy, and Company of Heroes is a nice exception. This is achieved by means of the Essence Engine, developed by Relic Entertainment. It supports all the modern technologies like advanced shader-based special effects, HDR, dynamic lighting and shadowing, normal maps, etc.
High-quality visuals never come out cheap. Let’s see how demanding Company of Heroes is.
It’s not that bad as we might have thought. Even last-generation single cards have acceptable speeds in 1280x1024 whereas the GeForce 8800 GTX easily delivers comfortable performance in 1920x1200, outperforming the Radeon X1950 XTX and the GeForce 7950 Quad SLI in two and three times, respectively.
Well, the GeForce 7950 Quad SLI system obviously has some problems here. It is only half as fast as the single GeForce 7950 GX2. It seems that the Quad SLI system turns 4x FSAA on correctly, but also switches to compatibility mode with only one out of its four GPUs active. We guess the current generation of GeForce 7950 Quad SLI will never really take off because the single GeForce 8800 GTX beats this complex, capricious and much more expensive quad-processor configuration in a majority of games.
The GeForce 7950 Quad SLI does better in this mode. It seems to be using SLI AA + SFR or SLI AA + AFR, although this operation mode has never been mentioned by the developer with respect to the GeForce 7950 Quad SLI. As a result, this system even beats the GeForce 8800 GTX in 1920x1200. But compare their minimum speeds, and you’ll see right away who’s the true winner here.
The GeForce 8800 GTX is on top again. In 16xQ mode it makes two out of the three resolutions playable, which makes this antialiasing mode practicable, as opposed to 16x SLI AA.
It’s hard to find out the winner here because all the top-end graphics cards have in fact the same speed, yet we can note that the Radeon X1950 XTX CrossFire is about 15% faster than the GeForce 8800 GTX in 1920x1200. Both cards are rather slow, but never slower than 35fps, which guarantees that you won’t have any slowdowns.
Like in many other cases, the GeForce 8800 GTX shows its worth in 8x FSAA mode. Its average performance is a mere 6fps lower in 1920x1200 in comparison with 4x FSAA mode and its min performance is only 2.2fps lower than the average frame rate. That is, there is almost no fluctuation of performance, which is an indication of a good reserve of speed.
In 16xQ CSAA mode the GeForce 8800 GTX behaves like in the previous test, but we’d recommend you to limit yourself to 1280x1024.
In spite of our expectations, the GeForce 8800 GTX doesn’t win in 3DMark05. It is as fast as the GeForce 7900 GTX SLI, though. The ATI Radeon X1950 XTX CrossFire is the closest to scoring 15.000 points – it’s got only 531 points less than that.
The top-end graphics solutions are all limited in speed by the CPU in 4x FSAA mode, but the GeForce 8800 GTX takes the lead when we switch to harder antialiasing modes. It’s natural as the new graphics card sports 32 TMUs, 24 ROPs and a 384-bit memory bus all of which are called for in the first test.
In the second test the above-mentioned advantages of the GeForce 8800 GTX add up with its unified architecture that allows to dynamically allot the streamed processors to perform shaders of different kinds. This only shows up in extreme antialiasing modes, though. In 4x FSAA + 16x Aniso mode the new card is even slower than the Radeon X1950 XTX CrossFire.
What we’ve said above about the second test can be applied to the third test, too. The GeForce 8800 GTX has excellent results in 8x MSAA and 16xQ modes, but loses to the Radeon X1950 XTX CrossFire when using ordinary 4x MSAA.
Thus, the final result is easily explained: the advantage of the Radeon X1950 XTX CrossFire tandem in the second and third tests in 4x FSAA + 16x Aniso mode also shows up at the default settings when there is no FSAA in use. 3DMark05 doesn’t suit too well for benchmarking solutions like GeForce 8800 GTX because this benchmarking suite doesn’t support Shader Model 3.0. Let’s check out the new card in the more modern 3DMark06.
It is the Radeon X1950 XTX CrossFire system that has the highest 3DMark06 score, outperforming the GeForce 8800 GTX by 199 points. Let’s examine this in more detail.
The GeForce 8800 GTX, GeForce 7900 GTX SLI and Radeon X1950 XTX CrossFire have similar results in the SM2.0 tests, yet the new graphics card is somewhat slower than the others.
The SM3.0/HDR tests are a more favorable environment for the GeForce 8800 GTX and it takes second place here. It scores 218 points less than the leading Radeon X1950 XTX CrossFire.
The GeForce 8800 GTX is unrivalled here because the first SM2.0 test is similar to 3DMark05’s first game test and needs a high fill rate from the graphics card. As our theoretical tests showed, the new graphics card from Nvidia has no rivals in this parameter.
The GeForce 8800 GTX is slower than the Radeon X1950 XTX CrossFire in the second SM2.0 test in spite of our using FSAA. We can’t name the reason precisely, but we suspect that low performance of the GeForce 8800 in some tests may be due to incorrect operation of the scheduler that is responsible for allotting the streamed processors of the G80 chip for shaders of different types.
In the first SM3.0/HDR test the GeForce 8800 GTX cannot take first place among graphics solutions that support FSAA and FP HDR simultaneously.
It’s the same in the second SM3.0/HDR test: the Nvidia card is ahead of the single Radeon X1950 XTX, but slower than the Radeon X1950 XTX CrossFire system.
Thus, 3DMark06’s overall scores seem to be well deserved. It is only in the first SM2.0 test that the senior model from Nvidia’s new graphics card family takes first place. The ATI Radeon X1950 XTX CrossFire wins the rest of the tests.
Nvidia’s desire to release its new-generation GPU ahead of the competitor led to unpleasant consequences. As we wrote earlier, there was a technical defect in the first batches of GeForce 8800 GTX. Cards from those batches came out with a resistor that had a wrong rating and the company had to withdraw them from the sales network. This problem didn’t concern the junior GeForce 8800 GTS.
As we found out, it is the R520 resistor, marked as 68C according to the EIA marking code table. It has a resistance of 49.900 Ohms, but there should have been a 25.500Ohm resistor in its place with a marking of 40C. Our samples of the GeForce 8800 GTX had the wrong resistor, too, but we corrected the resistance with an ordinary lead pencil. Nvidia claims that all GeForce 8800 GTX cards the users can buy have the right resistor on board, but at least a few users in the United States have reported they’ve bought defective samples of the GeForce 8800 GTX, as they claimed in some hardware forums (1, 2). Unfortunately, that was not the only issue with the GeForce 8800 GTX.
To our surprise, the GeForce 8800 GTX refused to work on certain mainboards. It started up without problems on an ASUS A8R32-MVP Deluxe (based on the ATI CrossFire Xpress 3200 chipset), but would not work at all on such mainboards as Intel Desktop Board D925XCV (Intel 925X, BIOS 0477, mainboard revision 1.0) and ABIT AN8 32X (nForce4 SLI X16, BIOS 13, mainboard revision 1.0). It was indeed queer with the latter mainboard because the graphics card would not be recognized by the system even if installed into the secondary PCI Express x16 slot with a GeForce 7600 GT in the primary one. ABIT claims they’ve found no incompatibility between the AN8 32X and the GeForce 8800 GTX. Perhaps there were some problems with the particular samples of the graphics card and mainboard.
We had no troubles with our EPoX EP-9NPA+ SLI (nForce4 SLI) mainboard, yet we had to give up the idea of benchmarking a SLI tandem built out of two GeForce 8800 GTX cards.
The current driver – ForceWare 97.02 – that supports the new series of graphics cards cannot be considered as stable and flawless.
Nvidia recommends using another driver for video playback tests because version 97.02 doesn’t support DirectX video acceleration. Another known defect of the current driver is that it is impossible to pass the pixel shader 2.x and pixel shader 3.0 checks from the Microsoft Display Compatibility Test, yet this is not a great problem for ordinary users because earlier generations of GPUs wouldn’t pass even more checks from the DCT. What is the most annoying thing about the current driver is that the GeForce 8800 GTX may have reduced performance with it. Or the speed of the card may fluctuate too much.
As we were benchmarking the card with ForceWare 97.02, we found out that performance of the GeForce 8800 GTX would fluctuate strangely and would often be much lower than expected. This showed up in 3DMark06 at first. We localized the problem eventually – the results of the card in 3DMark06 can be affected by System Registry entries left from earlier versions of ForceWare.
To avoid this problem, Nvidia recommends doing the following:
Although this helps solve the problem of fluctuating performance in 3DMark, it doesn’t solve it for other applications such as F.E.A.R. , The Elder Scrolls IV: Oblivion , etc. It means you have to reinstall Windows XP after which you can get the desired performance from your GeForce 8800 GTX.
The new Control Panel of the ForceWare 97.02 driver isn’t blameless, either. On one of our test computers it didn’t save the changes we made to the settings like FSAA and anisotropic filtering mode, and the VSync option. So, although we could check out the basic functionality of the GeForce 8800 GTX on the nForce4 SLI platform (EPoX EP-9NPA+ SLI), we couldn’t force the card to use its new FSAA modes on it. Besides other things, the new panel doesn’t seem to be very user-friendly. Its drop-down menus may be handier for inexperienced users, but they make the panel rather unwieldy, which will not be welcomed warmly by the fans of older Control panel version.
We were not alone in our troubles with ForceWare 97.02. Quite a lot of users who had bought the new card reported performance-related problems with their GeForce 8800 GTX/GTS on Nvidia’s official forum (1, 2, 3, 4). Moreover, some users of the new cards experienced not only slowdowns, but also saw image artifacts in games.
We didn’t see much of them, except for an incorrect fog in Half-Life 2: Episode One , but we had to reinstall the OS and some games a few times to obtain correct data and benchmark the GeForce 8800 GTX successfully, that’s why this test session took so long to get done. But we are absolutely sure that the numbers published above in the review are 100% correct and indicative of the real performance of the new graphics card.
GeForce 7900 GTX with FSAA 4x(left) and GeForce 8800 GTX with FSAA 8x (right)
The next-generation graphics card from Nvidia has passed through our tests and it’s time to do some summarizing.
Well, there’s no doubt the GeForce 8800 GTX has every reason to claim the championship title. It delivers fantastic performance and in most cases easily beats the best multi-GPU platforms of the last generation: Radeon X1950 XTX CrossFire, GeForce 7900 GTX SLI and GeForce 7950 Quad SLI. The number of benchmarks in which this graphics card with only one GPU on board proves to be faster than dual- or even quad-chip solutions is impressive indeed:
So, Nvidia triumphs in 14 out of 23 games we performed our tests in.
The GeForce 8800 GTX is especially good when you enable high-quality full-screen antialiasing like 8x MSAA or 16xQ CSAA. The developer has paid attention to this operation mode and endowed the new card with appropriate characteristics. Having 32 TMUs, 24 ROPs, and 768MB of fast graphics memory accessed across a 384-bit memory bus, the GeForce 8800 GTX feels at ease in modern 3D games even when you enable the most resource-consuming of antialiasing modes it supports, namely 16xQ CSAA.
The GeForce 8800 GTX is so strong that there’s generally no sense in using ordinary 4x antialiasing in resolutions like 1280x1024 and 1600x1200 because the card’s performance will be limited by the CPU, whereas 8x MSAA or 16x CSAA provide a much better-looking image.
Besides that, the GeForce 8800 family is free from the problem with anisotropic filtering quality that used to plague the previous generation of Nvidia’s graphics cards. This has a positive effect on the image quality, too.
Nvidia is triumphing, no doubt, but like any other cutting-edge product, the new graphics card is not free from drawbacks that mar its victory somewhat.
These are temporary problems, of course. New versions of ForceWare will correct the performance and stability issues, and gamers will have a $599 graphics card with unprecedented performance and excellent image quality. We hope Nvidia will make haste with that unless they want the reputation of such a promising product suffer.
Nvidia’s G80 processor incorporates as many as 681 million transistors while the use of 0.09-micron tech process has made the chip larger and its power consumption and heat dissipation, higher. GeForce 8800 GTX is thus a very complex and large graphics card, but the developer has managed to solve the problem of dissipating so much heat quite successfully.
However, before you purchase a GeForce 8800 GTX, make sure that:
You should also have a powerful enough CPU and enough of system memory and a monitor that supports resolutions of 1600x1200 and higher. As you could see in the tests, a CPU released less than a year ago doesn’t allow the GeForce 8800 GTX to reveal its potential fully in some games even if you enable full-screen antialiasing.
And finally, you should be aware that the ForceWare driver is not yet well optimized for the GeForce 8800 GTX.
Highs:
Lows: