by Alexey Stepin , Yaroslav Lyssenko, Anton Shilov
05/24/2007 | 03:39 PM
ATI Radeon HD 2900 XT graphics card undeniably has a lot of interesting capabilities, as we revealed in our article dedicated to the architecture of the Radeon HD 2000, the long-awaited sixth-generation of ATI’s Radeon. We also discovered several weak points of the new solution by the graphics product group of Advanced Micro Devices, but how will they affect real-world performance? We will find it out later in this article, but before that, let us remind you several key peculiarities of the Radeon HD 2000, certain facts about the Radeon HD 2900 XT as well as several findings about the new graphics card.
<%BANNER[article]%>ATI Radeon HD 2000 Architecture:
ATI Radeon HD 2900 XT Graphics Card:
With the review of the Radeon HD 2900 XT we decided to use our new Intel Core 2 Duo-based system with Windows Vista OS. The configuration of our new testbed is as follows:
Since we believe that the use of tri-linear and anisotropic filtering optimizations is not justified in this case, the graphics card drivers were set up in standard way to provide the highest possible quality of texture filtering.
ATI Catalyst:
Nvidia ForceWare:
For our tests we used the following games and benchmarks:
First-Person 3D Shooters
Third-Person 3D Shooters
RPG
Simulators
Strategies
Synthetic Benchmarks
Battlefield 2142 game does not pose a lot of load on geometry/vertex shaders, however, the game itself seems to put pretty heavy load pixel shader units of graphics processors.
All the participators here provide comfortable or nearly comfortable level of performance in resolutions up to 1600x1200. We cannot say that the GeForce 8800 GTS or the Radeon X1950 XTX deliver really good performance in 1920x1200 resolution, though, the speed is still nice.
ATI Radeon HD 2900 XT shows pretty illogical results in Battlefields 2142: on the one hand, its minimal framerate per second (FPS) is considerably higher compared to the GeForce 8800 GTS, whereas its average FPS is inline with the more expensive 8800 GTX.
Call of Juarez game title is full of high-definition rendering and pixel shaders, however, it does not look like to be full of geometry.
Only the GeForce 8800 GTX and the Radeon HD 2900 XT provide appropriate performance in 1280x1024 resolutions, while no other board tends to offer comfortable framerate neither in this, nor with other resolutions.
ATI Radeon HD 2900 XT is indisputably faster than Nvidia’s GeForce 8800 GTS, however it fails against the more powerful GTX model in 1600x1200 and 1920x1200 resolutions.
Far Cry is both pixel shader and texture intensive title. Moreover, Far Cry game tends to use as much as possible processing and system power for itself, which make it resemble id’s Quake 3 Arena of our times (unlike the aforementioned one, Far Cry still looks lovely three years after introduction).
Whenever you are going to get out of the cave and shoot your enemies near the deltaplan and then fly it to another shore while trying to get rid of mercenaries on the helicopter and the boat, any modern graphics card would fit in perfectly. The Radeon HD 2900 XT seems to be a little faster than others here, but not substantially.
Similar situations happens when it comes to “Research” level: if you are skilled enough, head-shots to the monsters here are nearly guaranteed, but in case you own the Radeon HD 2900 XT, everything is going to happen a little faster.
Crytek’s Far Cry was the first game to adopt high dynamic range (HDR) rendering and is still an example to follow…
Obviously, performance of modern graphics cards is so high that it is enough to play on the Pier level even on the Radeon X1950 XTX (well, perhaps, you will not be able to make headshots at the speed of lightning, but 50/40fps should be comfortable) and even in 1920x1200 resolution.
The results on the Pier level deserve attention: minimal frame per second figures tend to be similar on different graphics cards, but Nvidia’s GeForce 8800 GTX skyrockets even above itself in 1920x1200 resolution for an unknown reason. We do not know what limits performance here, but, definitely not render back ends of the new Radeon: even there are 16 of them, the new board tends to be faster than the predecessor (which has 16 as well) on a higher percentage than the difference between the clock-speeds of the boards.
Anyway, Nvidia’s GeForce 8800 GTX is the fastest here, while GeForce 8800 GTX and Radeon HD 2900 XT show similar results.
Research level shows no surprises: Radeon HD 2900 XT is slightly faster than GeForce 8800 GTS 640MB, while GeForce 8800 GTX sitting on the very top, but its minimal fps is constantly below those of AMD’s flagship offering. ATI Radeon X1950 XTX shows sufficient performance here too.
The well-known psychological thriller remains popular even these days, but what hardware do you need to play this game?
Well, you do have to own a nice graphics card to play in 1920x1440: even the GeForce 8800 GTX 768MB cannot guarantee 60 fps here. Nevertheless, we would say that this board can nearly deliver a pleasant level of performance: at least, framerate never drops below 25 fps. Other graphics cards only can guarantee flawless performance in 1280x1024.
Our today’s hero – the Radeon HD 2900 XT – competes nicely against the GeForce 8800 GTS 640MB, despite of the fact that its minimal fps is lower compared even to the Radeon X1950 XTX.
As usual, due to deferred rendering method used in the game’s engine, we cannot enable FSAA here.
The game is full of pixel shader effects, but it also contains quite a number of textures that need to be filtered with only 16 texture filtering units of the Radeon HD 2900 XT 512MB. As a result, the novelty can only demonstrate results similar to Nvidia’s GeForce 8800 GTS here: 8800 GTX is incomparable.
Not breaking any records, the Radeon HD 2900 XT is anyway competing successfully with the GeForce 8800 GTS. It is 5% slower at a resolution of 1920x1200, though. There can’t be any complaints about overall performance of the new AMD card – it is high enough for comfortable play in every display resolution with 4x FSAA enabled. This is largely due to the painstakingly optimized game engine that has a very moderate appetite for hardware resources as modern video games go.
Nvidia’s solutions seem to have no superiority in OpenGL games they used to enjoy before. The ex-flagship Radeon X1950 XTX easily maintains the same average speed as the GeForce 8800 GTS whereas the Radeon HD 2900 XT is some 15% ahead of its predecessor and is only inferior to the GeForce 8800 GTX.
At maximum level of detail and with enabled 4x full-screen antialiasing the new solution from AMD allows using resolutions up to 1600x1200 in this game. Its performance is near comfortable at 1920x1200, yet the Nvidia GeForce 8800 GTX is the only solution to offer the desired 60fps. The Nvidia card is far more expensive than the Radeon HD 2900 XT after all.
The game doesn’t support FSAA when you enable the dynamic lighting model, but loses much of its visual appeal with the static model. So, we benchmarked the cards in S.T.A.L.K.E.R. using anisotropic filtering only.
The Radeon HD 2900 XT fails this test completely. You can’t play normally using the dynamic lighting model even at 1280x1024. The new card being slower than the Radeon X1950 XTX, the reason is surely somewhere in the Catalyst driver rather than in the R600 GPU proper. On the other hand, recalling that the GeForce 8800 GTS 320MB often loses to graphics cards with 256MB of memory on board and that the developer hasn’t yet enabled FSAA for the dynamic lighting (the reason why we have to test S.T.A.L.K.E.R. with anisotropic filtering alone), we can suspect some defects in the game itself, not only in ATI’s driver. Anyway, you shouldn’t play S.T.A.L.K.E.R. on a Radeon HD 2900 XT 512MB as yet.
This seems to be another indicator of existing defects in the Catalyst driver for the ATI Radeon HD 2000 series. With all its superiority in technical characteristics, the results of the Radeon HD 2900 XT differ from those of the Radeon X1950 XTX, but only for the worse. The speed is high for playing the game in 1280x1024, but that’s not quite what you’d expect from a newest top-class 700-million-transistor graphics processor with a power draw of over 160 watts!
The Radeon HD 2900 XT isn’t brilliant in this game because the Next Generation Content mode requires high speed of texturing besides sheer computing power. There can’t be any competition with the GeForce 8800 GTX that has two times the number of TMUs and 1.5 times the number of ROPs that are clocked at just a slightly lower frequency. The new card from AMD accomplishes its main purpose of rivaling the GeForce 8800 GTS quite successfully, though, being no more than 11% behind in the worst case (in 1600x1200).
We try to get the best quality of graphics from each game, so we chose HDR in favor of FSAA (these features work incorrectly in this game when enabled simultaneously) and benchmarked the cards with anisotropic filtering only.
The Splinter Cell game series traditionally uses a lot of special effects with complex mathematical computations and Double Agent follows suit. The results are indicative of comparable computing capacities of the Radeon HD 2900 XT and GeForce 8800 GTX. These two cards are roughly equals, the AMD being somewhat better in 1280x1024 and the Nvidia in 1600x1200. Both cards provide maximum playing comfort. Unfortunately, the game does not support widescreen resolutions.
The current version of Gothic 3 does not support FSAA, so we benchmarked the cards using anisotropic filtering only.
The average speed of the Radeon HD 2900 XT is comparable to that of the Radeon X1950 XTX in every resolution, but the minimum of speed is somewhat higher due to the higher core frequency. Perhaps an insufficient amount of texture filter units prevents the new AMD architecture from showing its full potential. From a practical standpoint, the Radeon HD 2900 XT allows playing this game at 1280x1024 while the GeForce 8800 GTS provides a comfortable speed in 1600x1200, too.
Starting with version 1.04, the game allows using FSAA, but its support for HDR is still deficient. Therefore we tested the cards at the eye candy settings.
Like its predecessor, the AMD Radeon HD 2900 XT does not provide an acceptable speed in any of the tested resolutions. This may be the consequence of the new GPU design with only 16 TMUs and 16 ROPs that doesn’t match the NWN 2 game engine well. The game is very sensitive to the amount of those subunits as you can see by comparing the GeForce 8800 GTX and GTS, especially in high resolutions.
Like some of the games above, The Elder Scrolls IV: Oblivion doesn’t support FSAA and HDR simultaneously, so we tested it in HDR mode only.
Having a tremendous computing capacity, the R600 processor turns in good results in closed scenes of the game, although its results are lower than those of the GeForce 8800 GTX. Anyway, the Radeon HD 2900 XT is somewhat faster than the GeForce 8800 GTS and ensures a comfortable speed in every resolution, including the widescreen 1920x1200.
We’ve got a more complicated situation when testing the cards in open scenes. The need for high texturing speed is added to the demand for computing power. The Radeon HD 2900 XT can compute well and fast as is indicated by its min performance which is higher than that of the GeForce 8800 GTX even. But the new card’s average speed is indicative that the 16 texture filter units grouped into 4 texture processors are not enough – it is considerably lower than the average speed of Nvidia’s flagship (which is not a direct market opponent to the new AMD card, though). The Radeon HD 2900 XT compares well against the GeForce 8800 GTS: it can deliver more comfort in The Elder Scrolls IV , especially in 1920x1200 resolution.
The Radeon HD 2900 XT is almost as fast as the GeForce 8800 GTX in 1280x1024 and is a mere 5fps behind it even in 1600x1200, the average frame rates being as high as 100fps. The gap grows suddenly to 18% in 1920x1200, but the speed is anyway high enough for comfortable play and the new card is considerably better than the GeForce 8800 GTS in both average and minimum speeds in every resolution.
The game having a frame rate limiter, you should consider the min speeds of the cards in the first place. This parameter determines your playing comfort in Command & Conquer 3 .
The new $399 product from AMD is just as good as the Nvidia GeForce 8800 GTX in both minimum and average speeds in every resolution. This strategy is going to run smoothly on the Radeon HD 2900 HD.
We tested the game in the pure speed mode with enabled anisotropic filtering only, because it has problems when you turn on FSAA.
The ATI Radeon HD 2900 XT seems to be not much worse than the Nvidia GeForce 8800 GTX: from 6-7% in 1280x1024 and 1920x1200 to 10% in 1600x1200. The minimum speed of the new card is disappointing, though. It is not higher than 13fps in any of the tested resolutions. Even the ex-flagship Radeon X1950 XTX delivers a higher min speed, similar to the GeForce 8800 GTS. So, although the Radeon HD 2900 XT has a high average speed, it does not ensure comfortable gaming conditions, at least with the current driver.
Unfortunately, the Radeon HD 2900 XT is no faster than the Radeon X1950 XTX in Supreme Commander . We’ve seen in the other games that the new card from AMD is capable of more, so this must be due to some problem in the Catalyst driver. Hopefully this issue will be corrected soon.
The new card from AMD is successful in 3DMark05, scoring over 18.000 points. It outperforms the GeForce 8800 GTS as well as the more advanced and expensive GeForce 8800 GTX. But you should be aware that 3DMark05 defaults to 1024x768 resolution without full-screen antialiasing and it cannot reveal graphics architecture bottlenecks which are not directly related to performance of the shader subunits.
The fill rate and performance of the rasterization subunits are the theoretical bottlenecks in the Radeon HD 2900 XT architecture, so it makes sense to check out individual 3DMark05 tests at higher resolutions and with 4x FSAA.
We see it right away that the Radeon HD 2900 XT is no match to the G80-based cards in scenes that are sensitive to texturing speed because even the cut-down GeForce 8800 GTS has 24 TMUs with two texture filter units per each and 20 ROPs whereas the new card from AMD is limited to 16 units of each type, although clocked at a higher frequency. As a result, the Radeon HD 2900 XT can only surpass its predecessor, the gap between them being roughly equal to the difference in their core clock rates. The impressive memory bandwidth cannot help the new card, either, as the 1920x1200 results show.
The Radeon HD 2900 XT performs better in the second test. Not surprisingly as this test is not as heavy on the graphics card’s TMUs and ROPs as the first one. The second test needs high vertex shader performance and the new graphical architecture from AMD can deliver it as we made sure in our theoretical tests – it was much faster at processing geometry than the Nvidia GeForce 8800 series. The results may seem not very high from this point of view because the new card cannot reach the level of the GeForce 8800 GTS either due to the features of the test itself (3DMark05 is limited to version 2 shaders) or to some imperfections in the Catalyst driver.
Although limited to Shader Model 2.0, the third test is hungry for computing resources of the GPU as well as for texturing speed. The R600 has got an opportunity to strike back thanks to its 64 superscalar computing modules, each capable of executing 5 operations per clock cycle. The Radeon HD 2900 XT is still somewhat slower than the GeForce 8800 GTS, yet it passes through this test better than through the former two.
So, although winning in the default mode, the AMD Radeon HD 2900 XT cannot compete with the GeForce 8800 GTS, let alone with the GeForce 8800 GTX, in 3DMark05 if FSAA is enabled in popular display resolutions. We should make allowances for the age of this benchmark, though. It doesn’t use such features of modern GPUs as the support of Shader Model 3.0 and HDR lighting with floating-point color representation. 3DMark06 suits newest graphics cards much better.
The Radeon HD 2900 XT is roughly equal to the Nvidia GeForce 8800 GTX here. Considering their prices, this is a triumph for the AMD card, but where does this triumph come from? Let’s check out 3DMark06’s groups of tests as well as the results of individual tests.
It’s clear even in 1280x1024 that the small amount of TMUs and ROPs is the main problem for the Radeon HD 2900 XT. If it were not for this problem, the new card would easily overtake the GeForce 8800 GTX in the SM2.0 tests, but it has to be content with beating the 8800 GTS instead.
The new AMD card outperforms the GeForce 8800 GTX at last in the SM3.0/HDR tests. The difference is as small as 154 points, though. The R600 and G80 are indeed comparable when it comes to sheer computing power.
The results of the individual SM2.0 tests agree with what we’ve written above. The first test needs high texture-processing speed which is a weak point of the new Radeon as we know from our theoretical tests. The small advantage over the previous-generation top-end card from AMD is due to the difference in frequencies between the graphics core and, perhaps, due to the imperfect driver.
It must be due to this imperfect driver again that the new card delivers lusterless performance in the second test which puts an emphasis on geometry processing. As we know, the Radeon HD 2900 XT did especially well in our theoretical geometry-processing tests.
The GPU’s computing resources are demanded in the SM3.0/HDR tests and the Radeon HD 2900 XT feels more confident here, rivaling the GeForce 8800 GTS and almost matching the GeForce 8800 GTX in the second test.
Anyway, the AMD Radeon HD 2900 XT fulfils its main purpose – to compete with the GeForce 8800 GTS 640MB rather than with the monstrous GeForce 8800 GTX – quite successfully.
This benchmark makes its debut on our site. Rydermark was developed by the independent English company Candella Software with a specific purpose of demonstrating capabilities of modern graphics subsystems and evaluating their performance in new-generation games.
The benchmark is based on an engine developed for a future game. It supports Shader Model 3.0, 64-bit floating-point color representation, soft shadows, parallax occlusion mapping, realistic physics of liquids, and a lot of other features.
The test scene shows a race of speedboats through the canals of Venice, so the rendering of the water surface, which remains one of the most difficult tasks in computer graphics, is especially sophisticated. The benchmark displays results in abstract scores as well as in tangible frames-per-second.
To stress-test the modern graphics architectures from AMD/ATI and Nvidia we enabled 64x FSAA, 64-bit FP HDR and Parallax Occlusion Mapping.
The Radeon HD 2900 XT fails this round – the numbers are quite definite about that. Some problems with the driver for AMD’s new graphics card series may be the reason for the new card being slower than the Radeon X1950 XTX. Some problems in the benchmark itself may be another reason (Rydermark is currently beta 2, i.e. far from the final release).
This benchmarking suite may look somewhat odd among our traditional test applications. Its purpose is to test professional graphics cards under typical work conditions, i.e. not in games or in synthetic game engines like 3DMark’s, but in 3D modeling, CAD/CAM or scientific data visualization programs. We’ve added this benchmark to our testing program in response to multiple requests from our readers who use gaming-class graphics cards to run professional software due to the indecently high pricing of specialized graphics solutions. Although such solutions have lately been based on the same GPUs as gaming graphics cards, they come with special drivers that contain specific optimization. They also undergo complex certification and enjoy special support from the manufacturer. All of this is reflected in their price, of course.
Professional applications have specific features of their own. Particularly, they don’t use pixel shaders much, but instead require high speed and quality of texturing as well as high geometry-processing performance. Besides, such software often runs in the OpenGL environment or uses specialized APIs, so any optimizations for the gaming DirectX API are just not called for. SPECViewperf is a set of typical scenarios pertaining to professional applications, so we shouldn’t expect high performance from our games-oriented solutions. Anyway, let’s see what numbers we’ve got.
As the name suggests, the 3dsmax-04 scenario simulates a typical load on the graphics card in the Autodesk 3ds max suite. It uses geometrically complex models and from 5 to 7 light sources.
As you can see, the results of the Radeon X1950 XTX and both GeForce 8800 models are rather close, within 14-16fps, while the new Radeon HD 2900 XT is on the losing side. This must be due to its imperfect driver.
The frame rates are low. To get a higher frame rate, you should use a professional card from the Quadro or FireGL series.
The catia-02 scenario is for checking out a graphics card’s performance in the design and modeling system CATIA V5R12 developed by Dassault Systemes. It uses models containing from 1.2 to over 2 million vertexes, performs antialiasing verification, and changes model display modes on the fly, including a mode with textures and lighting.
Nvidia’s gaming solutions recede into the background, leaving top place to the humble Radeon X1950 XTX. The new Radeon HD 2900 XT has the worst result again. It suits poorly for working in CATIA, at least with the driver we used in our test session.
EnSight is a scientific data animation and visualization tool. It is frequently employed in physics research, particularly when studying the dynamics of liquids, the interaction of liquid and solid substances, etc. Each frame of the ensight-03 scenario from SPECViewperf 9.0.3 contains 3.2 million vertexes and some modeling modes make use of lighting, shading and texturing.
The Radeon HD 2900 XT is not as slower than the other cards in this test as it has been in the tests above, yet it still has the worst results. First place goes to the Radeon X1950 XTX again, which is closely followed by the GeForce 8800 GTX.
The LightScape Visualization System uses advanced physical lighting model algorithms to deliver fully realistic lighting in 3D scenes. The appropriate SPECViewperf scenario includes both wire models and models with smooth shading. Its geometrically most complex frame contains a total of 2.8 million vertexes.
The Radeon HD 2900 XT performs well enough yet is no better than the Radeon X1950 XTX despite all of its architectural improvements and innovations. The GeForce 8800 GTS is on the losing side, but the average speeds of the cards is only within 13-16fps.
This 3D modeling suite is used widely for creating special effects in movies and in games development, too. This SPECViewperf scenario uses a lot of models, the most complex of which has over 16 million vertexes and uses texturing. This is a heavy load even for a professional graphics card specifically optimized for such work.
Yielding over 25fps, the GeForce 8800 GTX regains the lead, but the Radeon X1950 XTX has a good result, too. The Radeon HD 2900 XT cannot reach a speed of 20fps even – due to its imperfect driver, again.
Widely employed in modern industry, the Pro/ENGINEER software is a tool for computer-aided design of mechanical components. This test scenario contains models with a complexity from 485 thousand to 5.9 million vertexes.
The Radeon HD 2900 XT is a loser again with a speed of less than 5fps. The GeForce 8800 GTX can yield a mere 11fps in this test, though.
Ordinary gaming graphics cards, even the most advanced ones, seem to be no match to professional graphics hardware. You can hardly hope to save your money on the graphics card you want to run professional software on.
Another computer-aided design system from Dassault Systemes, SolidWorks is a direct rival to Pro/ENGINEER. The sw-02 scenario simulates the operation of SolidWorks 2004 using two models: a car engine and a sports car.
The ATI Radeon X1950 XTX wins here like it has done in a number of earlier professional tests, notwithstanding Nvidia’s vast experience in developing professional graphics cards. Alas, the Radeon HD 2900 XT has nothing to boast about. Outperforming the GeForce 8800 GTS is the best it can do here.
Imitating the VisMockup visualization system, this test renders a complex model of a car with different shading modes containing from 5 to 11 million vertexes. This is one of the most resource-consuming tests in SPECViewperf 9.
And again the Radeon HD 2900 XT finds itself in last place while its predecessor Radeon X1950 XTX takes top position, considerably outperforming the GeForce 8800 GTX. Unfortunately, the overall performance level is as low as 2-4fps. It won’t be comfortable to work under such conditions.
NX 3 is a universal computer-aided design and development tool combining all stages of the development process: computer-aided design, computer-aided engineering, and computer-aided manufacturing. This SPECViewperf scenario contains a number of models of varying degree of complexity, up to 30 million vertexes with shading and lighting. This is going to be a hard trial for the graphics subsystem.
Having a well-balanced architecture that can only be let down by driver-related flaws, the GeForce 8800 GTX takes the lead here. It is followed by the GeForce 8800 GTS and Radeon X1950 XTX. The Radeon HD 2900 XT is lagging behind once again. This is indeed an ordeal of a test as you can judge by the extremely low performance of our gaming graphics cards.
Summing up the results of the professional tests we can come to only one conclusion which may sound disappointing for people who’d want to save money by using a gaming graphics card in professional applications. Gaming cards cannot deliver acceptable performance in 3D modeling, visualization, CAD/CAM and other professional programs because their drivers are written with other applications in mind and do not contain the optimizations that help professional graphics cards, based on the same GPUs, deliver much high performance under the same conditions. You can have some 25fps at best, but most often the speed of your gaming card is going to vary from 3 to 15fps. Not a comfortable speed at all.
Anyway, if you cannot afford a professional graphics card, the best choice among the gaming cards included into this review is Radeon X1950 XTX. It has won most of the SPECViewperf tests. The Nvidia GeForce 8800 GTX is second best, being a preferable solution for Maya and 3ds max.
ATI Radeon HD 2900 XT 512MB graphics cards left pretty ambiguous impression about itself. On the one hand, the graphics board is a good competitor for the GeForce 8800 GTS 640MB, but on the other hand its drivers deserve to be better.
The new graphics card has been awaited for many months and a lot of observers expected R600 with 700 million of transistors and 512-bit memory bus to leave Nvidia’s GeForce 8800 GTX in the dust. This does not happen and while the new Radeon HD 2900 XT can outperform the 8800 GTX model in certain cases, this board is really a competitor for Nvidia GeForce 8800 GTS 640MB.
Let’s summarize the cases when the novelty can actually beat Nvidia’s GeForce 8800 GTX, 8800 GTS and when ATI Radeon HD 2900 XT fails to beat its competitors:
* - first place
** - second place
*** - third place
Equal amount of asterisk means draw.
As it can be seen, there are only a few cases when ATI Radeon HD 2900 XT can compete against the GeForce 8800 GTX (Battlefield 2142 , Far Cry , Splinter Cell: Double Agent , Command and Conquer 3 ). In 5 cases out of 19 our today’s hero either demonstrates unacceptably low results, or comes behind GeForce 8800 GTS. The latter never shows inappropriately low results, but in 7 cases out of 19 it comes third.
Unfortunately, ATI decided to load the code-named R600 GPU with only 16 texture filtering units and 16 render back ends, which limits performance in a number of games, which is sad, as the chip features 512-bit memory bus and can definitely offer exceptional performance with FSAA enabled.
AMD’s driver team also could act much better: ATI Radeon HD 2900 XT has rendering issues with Half-Life 2: Episode One and performance problems with S.T.A.L.K.E.R. and some other titles.
It is clear that the Radeon HD 2900 XT with its current drivers cannot compete against Nvidia GeForce 8800 GTX or Ultra, which, by the way, cost substantially more, the latter costs ($829 and up) more than two Radeon HD 2900 XT graphics cards, in fact. But maybe ATI Radeon HD 2900 XT is the best graphics card for $399?
Well, it is either as fast, or faster than GeForce 8800 GTS 640MB in most cases, but it has performance issues with such games as S.T.A.L.K.E.R.: Shadow of Chernobyl , Neverwinter Nights 2 , Gothic 3 and probably some other. At some point those problems are going to be solved, but when – we do not know. So, performance-wise, ATI Radeon HD 2900 XT is a good solution.
In addition, AMD’s partners will include four free games with each Radeon HD 2900 XT graphics board: Half-Life 2: Episode 2 , Day of Defeat: Source , Team Fortress 2 and Portal . This increases overall value of the product to a gamer.
However, with sufficient performance and free games come power consumption of 161W as well as very noisy cooling system (at least, in case of our board) in addition to imperfect drivers, which, for example, do not allow to enable FSAA with edge-detect filter, something, which is projected to provide high image quality amid minimal performance drop. As a consequence of such driver limitation of the Radeon HD 2000, Nvidia’s GeForce 8800 can claim better image quality, as its CSAA works without an issue.
All-in-all, it is up to end-user to decide whether ATI Radeon HD 2900 XT is the best graphics card for $399 or not. It has a lot of advantages and the main one is truly extreme shader processing capabilities, but it also has a number of drawbacks.
Highs:
Lows: