by Tim Tscheblockov
10/23/2003 | 03:31 PM
On October 23, 2003 NVIDIA announced a new generation of DirectX9 compatible graphics processors. New NVIDIA GeForce FX 5950 Ultra (NV38) and GeForce FX 5700 Ultra (NV36) chips should strengthen the company’s positions in the competition with ATI, and are certainly aimed at competing with the recently announced RADEON 9800 XT and RADEON 9600 XT.
<%BANNER[article]%>The new graphics processors from NVIDIA, just like the newcomers from ATI, are based on the previous chips modifications, that is why you shouldn’t expect anything revolutionary from them.
For example, NVIDIA GeForce FX 5950 Ultra (NV38) is none other but an “overclocked” version of the previous generation NVIDIA graphics chips leader aka GeForce FX 5900 Ultra (NV35).
NVIDIA GeForce FX 5700 Ultra (NV36) is a more interesting chip, actually. The bigger part of it is similar to what we see in GeForce FX 5600 Ultra (NV31), but the vertex shader performance of the newcomer is considerably higher. Unlike the predecessor aka NV31, Nv36 features three vertex processors instead of two. It also supports DDR as well as DDR II graphics memory.
There are three graphics cards modifications, which will be based on the new NVIDIA NV36 and NV38 chips:
On ATI’s side there are two solutions announced this month, which should compete with NVIDIA in the graphics market:
In this review we are going to take a look at the fastest graphics cards models from ATI and NVIDIA, namely GeForce FX 5950 Ultra and RADEON 9800 XT. Better value and more mainstream solutions, such as GeForce FX 5700/5700 Ultra as well as RADEON 9600 XT are to come in the next articles. So, stay tuned!
In order to make the comparison of our today’s rivals more interesting I had to almost completely revise the set of gaming benchmarks. There finally appeared really worthy games capable of using the power of the newest graphics processors. All new games and synthetic benchmarks involved in our today’s test session use Pixel Shaders 1.x or 2.0.
In this case it is especially interesting to take a closer look at the top models of the today’s newest gaming graphics card families and estimate the efficiency of the new optimizing shader compiler, aka Unified Compiler, which was first officially included into NVIDIA drivers.
However, let’s better get started from the very beginning…
This solution will be represented by the NVIDIA reference board:
The PCB design of the new GeForce FX 5950 Ultra changed just a little bit compared with the PCB of GeForce FX 5900 Ultra, however, the cooling solution appears to be completely different:
The GPU cooling system of the NVIDIA GeForce FX 5950 Ultra looks very similar to the notorious FlowFX, although this is mostly an external similarity.
In the new cooling system the air is sucked inside through the holes in the graphics card bracket, then transferred through the plastic case to the fan and then is directed further to the heatsink. The air blows through the GPU heatsink ribs and then gets released inside the system case. The advantages of this cooling solution are evident: the graphics chip is always cooled down by the external cold stream of air, and there is no need to worry about the influence of the temperature inside the PC case on the graphics card stability and reliability at work.
The plastic air tube of the cooling system is fastened to the card with a few screws and hold the heatsink tightly pressed to the GPU surface thanks to a metal clip. This is actually very similar to Socket A cooler retention mechanisms, which also include a few tight fastening clips:
The fan installed on the NVIDIA reference card is pretty wide in diameter compared to that of the one used in FlowFX, and rotates at lower speed:
As a result, the cooling system of GeForce FX 5950 Ultra works much quieter compared to FlowFX and even noticeably quieter than a standard cooling system used on NVIDIA GeForce FX 5900 Ultra.
In terms of memory chips cooling, there is nothing new actually. The heatsink taking heat off the memory chips on the front side of the PCB got shaped a little bit differently, although the heatsink on the reverse side of it remained absolutely unchanged.
NVIDIA GeForce FX 5950 Ultra, like GeForce FX 5900 Ultra also features DVI-I, D-Sub and TV-Out ports, and also occupies two slots and requires additional power supply:
The graphics card is based on NVIDIA GeForce FX 5950 Ultra (NV38) working at 475MHz in 3D mode. Unfortunately, I cannot offer you and chip close-ups this time, because the heatsink is so firmly stuck on the thermal pad that I was concerned I could damage the chip when trying to remove it:
The card is equipped with 256MB of DDR graphics memory. These are the Hynix chips with 2.0ns cycle time:
The memory works at 950MHz frequency.
So, NVIDIA GeForce FX 5950 Ultra is none other but an overclocked version of NVIDIA GeForce FX 5900 Ultra and since the heat dissipation grew up as a result of that, the card definitely acquired a more efficient cooling system.
The new graphics card will hardly perform that much faster than the previous GeForce FX 5900 Ultra, because the GPU and memory frequencies in 3D mode have been increased only by 5.6% and 11.8% respectively.
However, NVIDIA has one more weapon ready to help NVIDIA strengthen its positions even more: the new optimizing shader compiler.
As is known, the DirectX 9.0 Pixel Shader performance of NVIDIA GeForce FX graphics processors is not as high as that of ATI RADEON 9500/9700/9800 chips. There are several reasons for that. First of all, lower Pixel Shader performance is the price you have to pay for higher flexibility of GeForce FX architecture (see our NVIDIA GeForce FX 5800 Ultra Review: New Technologies and Performance and NVIDIA GeForce FX 5900 Ultra (NV35) Review for more details on that). Secondly, it is no secret that DirectX 9 specifications were developed by Microsoft, which cooperated closely with ATI, shortly before the R300 chips series announcement. Therefore, it is not at all surprising that the DirectX 9 shaders are often “easier” for ATI graphics processors than for NVIDIA chips with a completely different architecture. As a result, it turns out that the shaders with the same input and output parameters and performed in the most optimal way on both: GeForce FX and RADEON 9X00, look differently.
In order to increase the Pixel Shader efficiency, NVIDIA developed an optimizing shader compiler, which takes into account all highs and lows of GeForce FX architecture and optimizes the process accordingly.
This is what the way the algorithm for “DirectX 9 Pixel Shader – GPU” looks like according to NVIDIA:
DirectX 9 Pixel Shaders are not used directly by the graphics processor. At first, DirectX translates the shader into a universal code and only then the graphics card driver compiles the code into pixel processor commands.
Exactly during the compilation of this intermediate code the new NVIDIA’s technology starts working. I am talking about NVIDIA Unified Compiler Technology, which is included into the driver called ForceWare. The optimizing compiler should take into account all GeForce FX peculiarities, so that it could create a chain of commands to be performance by the GPU as fast as possible.
For this purpose the compiler “takes apart”, analyzes and then “puts back together” the shader and can even rearrange the commands into pairs of texturing and mathematical tasks, reduce the number of involved time registers, reduce the overall number of commands by using only those of them, which are supported only by the GeForce FX hardware, etc.
The optimization criterion is the maximum processing speed of the output code. Besides, the output parameters of the initial and optimized shader should be in perfect coincidence. In other words, NVIDIA claims that the use of optimizing compiler eliminates any possibility of quality worsening.
This approach deserves all the applause. Firstly, the optimizing compiler can work with any shaders without any exceptions, and not only with those that are used in 3DMark093, for instance. Secondly, the company promises that the image quality will not suffer at all.
Well, when we compare the performance of NVIDIA GeForce FX 5950 Ultra with ATI RADEON 9800 XT, we will see if NVIDIA’s software developers managed to achieve all these goals.
The ATI RADEON 9800 XT graphics processor will be represented by the reference graphics card form ATI:
The PCB design of the new solution is pretty different from what ATI RADEON 9800 PRO used to have. The graphics chip and memory voltage regulator circuitry seems to have undergone the biggest changes.
The front side of the PCB is equipped with a new copper heatsink with a huge fan. On the reverse side of the card PCB the memory chips are covered with a copper heatsink plate for better heat take off.
Just like ATI RADEON 9800 PRO, the new card features DVI-I, D-Sub and TV-Out, and requires additional power supply. However, unlike NVIDIA GeForce FX 5950 Ultra, the ATI graphics card still occupies only one slot on the mainboard:
The cooling solution on ATI RADEON 9800 XT is a very quiet one, not any louder than a standard cooler of a RADEON 9800 PRO based graphics card.
The card is built on ATI RADEON 9800 XT graphics processor working at 412MHz core frequency:
The card feature 256MB of onboard DDR SDRAM graphics memory in Hynix chip with 2.5ns cycle time:
The graphics memory works at 730MHz (365MHz DDR).
The graphics card based on ATI RADEON 9800 XT boasts about 8.4% faster graphics core than the predecessor, and about 7.3% faster graphics memory, that is why we do not expect RADEON 9800 XT to be too far ahead of the RADEON 9800 PRO, to tell the truth.
Or maybe we should? ATI suggests that you should increase the performance of your RADEON 9800 XT with a few dynamic overclocking tricks. Let’s find out what they are!
As soon as Catalyst driver version 3.8 came out, you could notice not only a few smaller enhancements in the 3D graphics page, but also a new SMARTSHADER Effects page where you could enable post-processing of the 3D application images, and an OVERDRIVE page, which we would like to draw your attention to:
When you enable the Overdrive function, the driver automatically increases the ATI RADEON 9800 XT graphics processor frequency. If the core is no warmer than 52oC (the R360 chip features a built-in thermal diode), the frequency rises up to 432MHz. If the temperature increases even more but doesn’t reach 62oC, the new frequency will make 418MHz. When the graphics chip gets warmer than that, the frequency will automatically drop down to the nominal value.
In order to avoid system reboots in case the graphics card “freezes dead”, because of the graphics chip overheating, for instance, you can enable VPU RECOVER feature:
In this case if the core stops responding to the driver requests, the driver will reinitialize the graphics processor.
So, 418MHz and 432MHz during dynamic overclocking. Is this a significant frequency increase or a small one? Relative to the nominal GPU frequency of 412MHz, these values are equal to 1.5% and 4.9% respectively.
Well, overclocking is a useful thing, of course, but the maximum of 5% is too little, really. I think that if there are such nice hardware monitoring options, the ability to restore the default nominal frequencies in case of overheating and reinitialize the VPU in case of emergency, the frequencies could really be increased much farther beyond 5%. Especially, since we know that R350 chips and evidently, R360 chips too, boast excellent overclocking potential and 20% frequency increase, which turns into 40% once you apply extreme overclocking methods is far not the top of their ability (see our article called Extreme Overclocking Experience: NVIDIA GeForce FX 5900 Ultra against ATI RADEON 9800 Pro).
When we tested ATI RADEON 9800 XT, we decided not to use the OVERDRIVE function, although we liked the idea of dynamic overclocking a lot. As a result, we tested this graphics card at the nominal frequencies. Why? Because those of you who plan to overclock this card, will hardly bee happy with a 5% frequency increase. And those who simply select a specific menu item in the control panel will hardly notice any real performance improvement.
ASUS Company announced their cooperation with ATI the same day as the new RADEON 9800 XT and RADEON 9600 XT were released. As a result, there appeared a totally new graphics card family from ASUS based on the ATI chips, which included not only the brand new models, but the entire product range starting with RADEON 9200 and finishing with the latest RADEON 9800 XT.
In our test session the ATI RADEON 9800 XT chip will be represented by a product from ATI’s new partner aka ASUS RADEON 9800 XT/TVD:
The carton box with the graphics card is simply huge. To really feel how big it is, take a look at the box shot with the open top: there is a transparent window where the graphics card actually is:
The PCB of the new ASUS solution is slightly different from ATI’s own reference design:
The cooling solution consists of a dual-fan cooler covering the graphics chip and the memory on the front side of the PCB and a copper plate taking the heat off the memory chips on the reverse side of the card PCB:
By the way, the heatsink on the front side of the PCB is not made of solid copper, as you might have thought at first sight. This heatsink basically has a copper plate with a heat pipe stuck to it and aluminum heatsinks screwed to the plate with small screws:
This cooler seems to be a really efficient one, however, it is noticeably noisier than the reference card cooler.
The graphics card is based on ATI RADEON 9800 XT video processor working at 412MHz:
And features 256MB of graphics DDR SDRAM in Hynix chips with 2.5ns cycle time:
The graphics memory of ASUS RADEON 9800 XT/TVD works at 730MHz (365MHz DDR).
Like any other RADEON 9800 PRO or 9800 XT based graphics card, ASUS RADEON 9800 XT/TVD requires additional power supply. The card is actually more functional than ATI’s reference solution, because it features not just a Video-Out but also a Video-In. the ports on the graphics card bracket are absolutely the same as those of the reference card, but the Video-In/Out combo on the ASUS card also uses the pins responsible for the Video-In:
The Video-In function, namely, video signal decoding, is implemented via the ATI Rage Theater chip installed on the reverse side of the PCB:
By the way, this is one of the differences between ASUS RADEON 9800 XT/TVD and the reference design from ATI: judging by the looks of the ATI’s reference card I can say that they probably implied to use a more up-to-date Rage Theater 200 chip with a different package type.
The package of ASUS RADEON 9800 XT/TVD contains everything you might need to work with a graphics card: a set of brochures, CD-disks with the drivers and utilities including PowerDirector 2.55 ME and MediaShow SE 2.0 from Cyberlink , Ulead Cool 3D SE 3.0 and Ulead Photo Express 4.0.
To connect the additional peripherals for work with Video signals, ASUS provides a special adapter with two RCA connectors and two S-Video connectors:
In conclusion to this part of our coverage I would like to say that the graphics card from ASUS is equipped with very well-thought hardware monitoring features. Their brand name SmartDoctor utility informs the user about the graphics chip and memory temperatures and voltages, tracks AGP voltage and fan rotation speeds and allows setting some limit values for them, notifying the user once they get surpassed. Besides that, SmartDoctor also allows setting clock frequencies for the VPU and graphics memory and resets them to defaults automatically in case of overheating. At last, if quiet operation is what you value most, there are two ways to control the rotation speeds of the cooler fans: by setting a certain rotation speed or by enabling automatic fan rotation speed management depending on the core temperature.
Unfortunately, you will be able to enjoy this whole features set only if you have an ASUS graphics card. On ATI’s graphics card the SmartDoctor utility will not work at all. However, the standard OVERDRIVE function from Catalyst 3.8 control panel, which we have already discussed above, also refused to work on ASUS RADEON 9800 XT/TVD card.
As for the performance of ASUS RADEON 9800 XT/TVD compared with that of the reference card from ATI, there are no surprises here: when working at nominal frequencies, both cards demonstrate almost the same performance results.
Well, let’s get down to the most exciting part: the benchmarks!
Our testbed was configured as follows:
We used the following software:
The anisotropic filtering settings and texture level of detail were set to Quality for all graphics cards tested.
Unreal Tournament 2003 is the only benchmark left from the old gaming test package, which we have bee using on our site. During the tests in this game I used 32bit modes and maximum graphics quality settings: Texture Detail - Highest, World Detail - Highest, Character Detail - Highest, Physics Detail - High, Character Shadows - ON, Dynamic Lighting - ON, Detail Textures - ON, Projectors - ON, Trilinear Filtering - ON, Decals - ON, Coronas - ON, Decals Stay - High, Foliage - ON, Use Blob Shadows - OFF.
All test were run in my own demo-record created in Unreal Tournament 2003 DM-Inferno level:
So, here come the results:
ATI and NVIDIA graphics cards run almost neck and neck:
When we enable full-screen anti-aliasing, NVIDIA GeForce FX 5950 Ultra and NVIDIA GeForce FX 5900 Ultra start dashing ahead of the competitors from ATI, especially as the resolution increases. They manage to win due to higher memory bus bandwidth, which is more important for performance at high screen resolutions.
Anisotropic filtering was forced not in the driver but in the game settings. When this function was forced in the driver, the ATI based graphics card wouldn’t apply tri-linear filtering to base textures of Unreal Tournament 2003, but only anisotropic filtering, even though the driver was forcing the Quality mode. This is the result of ATI’s anisotropic filtering “optimization”, when even in the Quality mode anisotropic and tri-linear filtering are enabled only for the zero texturing stage, and for all additional textures only anisotropic filtering is performed.
NVIDIA’s chips also boast some “optimizations” for tri-linear and anisotropic filtering algorithms. For example even in the Quality mode GeForce FX 5950 Ultra and GeForce FX 5900 Ultra do not actually use the fully-fledged tri-linear filtering, but a “mixture” of tri-linear and bi-linear filtering methods (see our NVIDIA GeForce FX 5800 Ultra Review and NVIDIA GeForce FX 5900 Ultra Review for more details).
However, this trick didn’t help NVIDIA chips here: ATI RADEON 9800 XT and 9800 PRO manage to defeat them due to faster though less quality anisotropic filtering algorithm.
Enabling FSAA doesn’t change the picture at all: ATI RADEON 9800 XT and RADEON 9800 PRO are still ahead.
So, Unreal Tournament 2003 benchmark didn’t reveal anything new: the performance in older games has long been studies inside out, so the results of Unreal Tournament 2003 could really be predicted with pretty high precision.
Homeworld 2 is a classical 3D strategy game repeating the well-known predecessor, Homeworld. There are no outstanding special effects in this game, however, the scenes are really beautiful all the way.
Homeworld 2 depends on the CPU really severely, so that you can actually see the performance differences caused by the use of different graphics cards only in the rare “moments of peace” between the battles, in relatively “calm” missions or in the intro-movies for the new levels. All the rest of the time the engine is busy calculating the enemy’s AI, gaming physics and logic, so that the benchmark results are limited by the CPU performance a lot.
Nevertheless, the game also proved to be loading graphics cards quite heavily. Therefore, being a real Homeworld fan, I couldn’t resist the temptation and test the cards in one of my favorites. :)
The graphics quality settings were set to maximum. The dynamic control of models complexity (in other words, making the models simpler as the distance to them increases) was disabled. The tests were run in two modes: with and without shadows. Anisotropic filtering and anti-aliasing were disabled, because the former doesn’t make sense for a game like that and the latter doesn’t work on NVIDIA graphics cards with driver version 52.16. the tests were run in a scripted scene in the beginning of the second game level with the help of Fraps utility.
Here are the results:
The graphics cards on ATI’s chips are ahead of the NVIDIA’s rivals. Homeworld 2 is a convenient application for them, because most objects in the scene use maximum one texturing layer, which is ideal for ATI chips with 8 pixel pipelines with one texturing unit in each.
When the shadows were enabled the situation changes: in all resolutions up to 1600x1200 NVIDIA solutions are ahead. Probably the gaming engine uses a stencil buffer for shadows calculation and NVIDIA chips build stencil buffers twice as quickly, involving 8 pixel pipelines.
ATI chips seem to be calculating the shadows in a different way. Firstly, the benchmark results show that the performance difference in various resolutions is very small, i.e. the graphics card performance is no bottleneck here. Secondly, the external shadows by ATI chips look a little different in Homeworld 2.
So, with the maximum quality settings and with enabled shadows NVIDIA graphics cards win in Homeworld 2 test.
Halo is a classical 3D-action game, very similar to Unreal and Unreal 2. the game was first developed for X-Box gaming console based on NVIDIA chip. However, then they ported the gaming engine to the PC platform and as a result it needed to be seriously modified. The Halo graphics engine uses pixel shaders 2.0, 1.4 and 1.1 for various surfaces and effects.
Halo features special built-in options for graphics cards performance adjustment. When you start the game with the “-timedemo” key, Halo displays a sequence of scripted scenes as fast as possible, and them shows you the average fps rate for all of them. Unfortunately, scripted scenes used here appear much easier for the graphics cards compared to the workload during real gaming experience. However, these results still give us some idea of the graphics cards performance in this game.
Halo tests were performed with forced pixel shaders 2.0 and with maximum graphics quality settings.
We didn’t run any tests with enabled full-screen anti-aliasing, because Halo doesn’t support this function.
Here are the obtained results:
Well, the results of applied optimizations seem to be more than evident. NVIDIA graphics cards perform as fast as ATI based ones. However, they owe these good results not only to the shaders optimizing compiler. On NVIDIA GeForce FX graphics cards, the Halo graphics engine uses half-precision pixel shaders 2.0, which provides the NVIDIA GPUs with some performance advantages.
However, I didn’t reveal any issues caused by these pixel shaders optimizations or any calculations precision worsening. The only remark here is that sometimes some objects of the scenes appear to have no fog around them.
When we enabled anisotropic filtering, ATI performed somewhat batter, which I have actually expected.
So, the new NVIDIA and ATI chips perform almost similarly in Halo. Only the enabled anisotropic filtering provides ATI RADEON 9800 XT and 9800 PRO with some performance advantage over the rivals.
Tron 2.0 is a “fresh” 3D action game based on the story from the Tron movie, one of the first movies where they used computer graphics effects. The gaming environment is the insides of the PCs, servers, routers and the like, as the developers see them. And these guys really seem to have excellent imagination :)
The game is a very unusual and beautiful one. However, even though the LithTech engine from Monolith requires installed DirectX9, it only uses DirectX8 pixel shaders. The most remarkable effect used in Tron 2.0 game is the multi-color halo around the brightest objects. NVIDIA and Monolith together developed this special effect for Tron 2.0 game, and without it the game lose at least half of its unique charm.
For tests in tron 2.0, I selected a scripted scene from City Hub level and used Fraps for performance measurements:
The graphics quality settings in the game were maximum. The only exception was anisotropic filtering: we disabled it in the game settings, because the driver anyway forced it when necessary.
Here come the results:
ATI based graphics cards are ahead of the competitors, because they are in a bit better situation: most Tron 2.0 objects are created with a single texture layer, and ATI chips are known to be performing the DirectX8 pixel shaders faster than NVIDIA GeForce FX GPUs.
Despite the enabled full-screen anti-aliasing ATI chips retain the leadership.
Enabled anisotropic filtering provides ATI chips with an additional advantage over the rivals, as they are using a fast though less quality algorithm.
So, even though NVIDIA took active part in Tron 2.0 development and the into-message says “The way it’s NVIDIA meant to be played”, GeForce FX 5950 Ultra failed to outperform ATI RADEON 9800 XT.
Tomb Raider is Tomb Raider, there is nothing I could add here. The latest Tomb Raider incarnation aka Angel of Darkness (TRAOD) acquired the support of modern graphics processors and such exciting special effects as light refraction in streams of hot air, various reflections on “complex” surfaces and in the water, blurring of objects beyond the camera focus, etc.
When we tested NVIDIA and ATI based graphics cards in Tomb Raider: Angel of Darkness, I used identical settings, which are inserted during profile PS 2.0 activation. I actually made two changes to these settings. First, I disabled video synchronization and second, I disabled Pixel Shader 2.0 Shadows, because the NVIDIA based graphics cards went through emergency game termination every time I tried enabling this option.
I used special integrated tools of the Tomb Raider: Angel of Darkness to measure the performance of the testing participants, namely I played the preliminarily recorded scene. I have made a total of two records: one on paris2_3 level and the other one on paris5_4 level.
The first record uses fewer pixel shaders. As you can see from the screenshot below, they only apply blurring to the distanced objects located outside the camera focus:
So, first come the results for paris2_3 record:
ATI RADEON 9800 XT and RADEON 9800 PRO leave not a single chance to NVIDIA chips.
However, as soon as we enable FSAA, NVIDIA GeForce FX 5950 Ultra and GeForce FX 5900 Ultra manage to get ahead of their rivals. As you can se from their results in higher resolutions, even though they are limited by the pixel shader processing speed, they hardly lose anything when the FSAA is enabled. While ATI RADEON 9800 XT and 9800 PRO slow down quite tangibly because of that.
Enabled anisotropic filtering allows ATI chips defeat the competitor, as we have expected.
In the heaviest mode NVIDIA graphics cards still manage to retain the leadership.
In paris5_4 record they also try to model light refraction in hot streams of air. Besides, the scene is heavier loaded than the previous one from the start:
So, here are the results obtained in paris5_4 scene:
In the harder scene the victory moved even more towards ATI based testing participants.
Enabled full-screen anti-aliasing doesn’t deprive ATI from the leadership in this game.
Forced anisotropic filtering, as always contributes to ATI’s triumph.
Even in the heaviest mode ATI graphics cards are still ahead of all.
So, the only conclusion I can make here is not very promising for NVIDIA based solutions. Even with the new driver the NVIDIA based graphics cards cannot boast high gaming performance in gaming, such as Tomb Raider: Angel of Darkness, which are actively using Pixel Shaders 2.0
To Tell the truth, I tend to consider Aquamark3 test set not a synthetic graphics cards performance benchmark, but an excellent demonstration of the potential implied in the new Massive Development’s gaming engine.
The test consists of 9 episodes. Each of these episodes demonstrates one of the technologies used by the engine. Despite that, there are no significant differences between all 9 scenes, which I personally regard as a certain disadvantage, because it makes the test not “synthetic” enough.
Taking this fact into account I would like to offer you the average performance results for all nice benchmarks, because it doesn’t make much sense to analyze each of them separately. The scenes level of detail was set to the maximum throughout the entire test session:
In the lightest mode without FSAA and anisotropic filtering, NVIDIA based graphics cards appeared a little ahead of the competitors.
When the FSAA is enabled, ATI solutions manage to outperform the competitors. It is probably not the memory bus bandwidth that matters here the most, but the pixel shaders processing speed and the texturing speed.
Enabled anisotropic filtering again pushes ATI forward.
Here I would like to stress that when you involve configuration files and enable anisotropic filtering, Aquamark3 will not use tri-linear filtering, even though you will see it enabled in the settings.
In the heaviest mode, ATI chips manage to retain their advantageous position.
Well, the last of our today’s tests again declared the failure of NVIDIA’s solutions.
So, the announcement of NVIDIA GeForce FX 5950 Ultra didn’t help NVIDIA to become a leader in the High-End gaming market. However, no one had actually expected that to happen: “slightly overclocked” GeForce FX 5900 Ultra can hardly be able to compete with ATI RADEON 9800 XT in new games, which are using DirectX9 Pixel Shaders.
However, NVIDIA not just launched the new GeForce FX 5950 Ultra and 5700 Ultra. They also released a new ForceWare driver featuring an optimizing compiler for DirectX9 pixel shaders. Of course, the new driver will not be able to eliminate limitations imposed by the GPU architecture, but it definitely makes NVIDIA GeForce FX 5950 Ultra look not so helpless against the background of ATI RADEON 9800 XT.
Nevertheless, ATI RADEON 9800 XT remains the performance leader in contemporary games and the today’s test session really proves this point. Taking into account that the recommended price of ATI RADEON 9800 XT is $499, which is exactly the same as the price of NVIDIA GeForce FX 5950 Ultra, there shouldn’t be any questions about the choice of a High-End graphics card any more.
However, the major battles will actually take place far from the High-End segment. Together with the NV38 GPU, NVIDIA announced NV36, which will be available in the market in two basic modifications: GeForce FX 5700 and GeForce FX 5700 Ultra. These graphics processors are targeted for the mainstream segment and should be selling for $149 and $199, as recommended. Well, I have the feeling that this is going to be a really exciting competition with ATI RADEON 9600 and 9600 XT.
We are going to review the new ATI and NVIDIA solutions targeted for the mainstream market in the nearest future, so stay with us!