This is a copy of the reference card with just another cooler. There is even a cut on the memory sink below on the right of the cooler (that cut was made for the reference cooler of the parallelogram shape).

Now comes the cooler.

Sapphire Atlantis RADEON 9800 PRO 256MB Ultimate Edition
The cooler consists of two massive aluminum heatsinks with a copper gold-plated pipe with low-boiling liquid between them. This pipe carries heat from the front heatsink to the rear one. There is thermal grease between the pipe and heatsinks where they touch each other. The mounting system is very reliable and consists of screws near the chip and buckles between the sinks over the card.

That's nice that the sinks are placed quite high, and it's guaranteed that the north bridge coolers located close to the AGP slot won't be an obstacle on the way of such card.

This is how the card looks in the mainboard.

And this is after 10 hours in 3D. The sink got red :-) (it's a joke).

You can see that if a PCI card has components on its backside it can't be inserted into the first slot; besides, such card would prevent proper ventilation.

The card ships in the retail package.

Testbed and drivers

Testbed:

• Pentium 4 3200 MHz based computer:
• Intel Pentium 4 3200 MHz CPU;
• DFI LANParty Pro875 (i875P) mainboard;
• 1024 MB DDR SDRAM;
• Seagate Barracuda IV 40GB HDD;
• Windows XP SP1; DirectX 9.0a;
• ViewSonic P810 (21") and ViewSonic P817 (21") monitors.
• ATI drivers v6.368.

VSync off, S3TC off in applications.

Cards used for comparison:

• Reference card NVIDIA GeForce FX 5900 Ultra (450/425 (850) MHz, 256 MB, driver 44.90).

Test results

Before we start examining 2D quality, I should say there are no complete techniques for objective 2D quality estimation because:

1. 2D quality much depends on certain samples for almost all modern 3D accelerators;
2. Besides videocards, 2D quality depends on monitors and cables;
3. Moreover, certain monitors might not work properly with certain video cards.

With the ViewSonic P817 monitor and BNC Bargo cable the card showed excellent quality at the following resolutions and clock speeds:

In general, 2D quality is excellent (but of this sample and in these conditions).

Test results: cards' performance

Conventional signs: ANISO 8xP - Anisotropic 8x Performance (earlier it was called Balanced), ANISO 8xQ - Anisotropic 8x Quality, ANISO 16xQ - Anisotropic 16x Quality.

Test applications:

• Return to Castle Wolfenstein (MultiPlayer) (id Software/Activision) - OpenGL, multitexturing, ixbt0703-demo, test settings - maximum, S3TC OFF, the configurations can be downloaded from here
  
• Serious Sam: The Second Encounter v.1.05 (Croteam/GodGames) - OpenGL, multitexturing, ixbt0703 demo, test settings: quality, S3TC OFF
• Quake3 Arena v.1.17 (id Software/Activision) - OpenGL, multitexturing, ixbt0703 demo, test settings - maximum: detailing level - High, texture detailing level - #4, S3TC OFF, smoothness of curves is much increased through variables r_subdivisions "1" and r_lodCurveError "30000" (at default r_lodCurveError is 250 !), the configurations can be downloaded from here
• Unreal Tournament 2003 v.2225 (Digital Extreme/Epic Games) - Direct3D, Vertex Shaders, Hardware T&L, Dot3, cube texturing, default quality
  
• Code Creatures Benchmark Pro (CodeCult) - the game that demonstrates card's operation in DirectX 8.1, Shaders, HW T&L.
  
• Unreal II: The Awakening (Legend Ent./Epic Games) - Direct3D, Vertex Shaders, Hardware T&L, Dot3, cube texturing, default quality
  
• RightMark 3D v.0.4 (one of game scenes) - DirectX 8.1, Dot3, cube texturing, shadow buffers, vertex and pixel shaders (1.1, 1.4).
  

If you want to get the demo benchmarks that we use let me know by e-mail.

There are a lot of materials on the Net that the trilinear filtering doesn't work properly in UT2003 in NVIDIA's cards (see http://www.beyond3d.com/, HardOCP etc.). From time to time we will touch upon this issue.

Quake3 Arena

Well, they are almost equal. In comparison with the Quality level the RADEON 9800 has a higher speed, but the quality of this function is higher in case of NVIDIA. If we compare it with the Performance level, where NVIDIA has its own downsides like inferior trilinear filtering and anisotropy on far objects, RADEON 9800 will lose its leadership.

Now quality of these functions. First comes the pure trilinear filtering (without anisotropy).

RADEON 9800

  

The quality is OK. Now with the anisotropy 16x Quality:

RADEON 9800

  

Now the high-level trilinear filtering, all functions work entirely in the OpenGL .

Serious Sam: The Second Encounter

The RADEON 9800 falls behind in the heavy modes, and comes ahead in the light ones. If we compare the Quality/Quality anisotropic levels, the Sapphire comes forward. But in general, they go on a par taking into account high quality of the RADEON 9800 in the AA4x mode compared to the GeForce FX 5900.

Return to Castle Wolfenstein (Multiplayer)

Here the RADEON 9800 takes the lead in most cases. But if we compare the Performance (NVIDIA) to Quality (ATI) the GeForce FX 5900 goes ahead due to its anisotropy optimization and simpler trilinear filtering.

Code Creatures

In the v44.90 of NVIDIA the cheats for this benchmark are removed. Plus, normal anisotropy. The RADEON 9800 has a solid advantage! Here are also results of the FX5900 with the Anti-Detect, that proves that the cheats are absent. Unfortunately, the Anti-Detect works only in the Detonator up to v44.65 inclusive; plus, it has problems with anisotropy, that is why we can compare their performance with AD only without anisotropic filtering. By the way, the developer of Anti-Detect Aleksei Nikolaichuk AKA Unwinder is thinking why the NVIDIA driver doesn't turn on anisotropy in this test when the Anti-Detect is on. Obviously, the latter works but the driver is not aware that this particular benchmark is started. Why doesn't the Detonator enable this function? The question is open.

Unreal Tournament 2003

There is an approximate parity again. With the optimizations disabled the speed of the FX5900 sharply falls down, and the RADEON 9800 has a great advantage. We have already discussed why NVIDIA uses optimizations in UT2003 and doesn't Unreal II (the game on the same engine). The examination with the Anti-Detect on the ATI drivers showed that no cheats are used in any game tests (3DMark is not used anymore). That is exactly why we compare the FX5900 with Anti-Detect with the RADEON 9800 without AD.

Now the quality issue. In the UT2003 we can enable trilinear filtering and choose an anisotropy level. Let's turn on only the trilinear filtering (High Quality of ATI and NVIDIA). The MIP levels are colored from the console with the variable "FirstColoredMIP 1".

RADEON 9800

  

FX 5900

  

FX 5900 with Anti-Detect

  

One more sin of NVIDIA. Isn't it the gist of their optimizations? Although the Quality preset is used, the trilinear filtering is not of the high level. The Anti-Detect reveals the real situation. Judging by the speed difference, the trilinear filtering alone can't cause such difference. But in the game the difference between the simplified and normal trilinear filtering isn't noticeable, the borders between the MIP levels do not catch the eye.

Now let's force anisotropy 16x and 8x at Quality and Performance (in the game itself the game is disabled).

RADEON 9800 Quality

  

RADEON 9800 Performance

  

FX 5900

  

FX 5900 with Anti-Detect

  

One more surprise! As to NVIDIA, it was clear even in the previous test that they use simpler trilinear filtering which was proved with the Anti-Detect. And what about ATI? I've got a bad feeling... There is no trilinear filtering at all! Then what?

Now let's enable in the game with ut2003.ini and disable in the drivers.

RADEON 9800

  

FX 5900

  

Now the RADEON 9800 uses the normal trilinear filtering! It turns out that the driver detects a game, asks it whether it knows about anisotropy, and if the game says it doesn't "know how", the anisotropy forcing is successful. And if the game "knows how", the driver "overcomes" settings in the game and sets user's ones. Or it can give the control over this function to the game, and until the game enables its optimization it doesn't work. Both variants are logic, though the second one is strange because - why forcing then? But there is the third variant that works indeed: the ATI driver detects a game, asks for the anisotropy level enabled, and if the function is off, it sets a lower level for the trilinear filtering. May be a cheats?

No. ATI engineer tells me, and forum of Beyond3D has the same information. ATI doesn't detect the application, this is just how their control panel AF works. In Quality mode only texture stage 0 has Trilinear, all other stages have Bilinear - so with multitexturing you are getting a Trilinear and Bilinear blend. This is occurring in all cases with the app controlled AF, but you only "see" it in UT2003 because of the nature of the textures. Any application that has AF controls, as UT2003 does, will receive full Trilinear on all layers if the application requests it.

The FX 5900 doesn't show normal trilinear filtering without the Anti-Detect either.

Now we leave the anisotropy enabled in the game and force it in the drivers.

RADEON 9800

  

FX 5900

  

The NVIDIA driver didn't understand what's they want from it and disabled the true trilinear filtering at all :-) ATI driver makes the bilinear.

In general, everyone is guilty including the game developers.

• ATI: the normal anisotropy and the trilinear filtering work only when they are enabled in the game itself; for this purpose you must change ut2003.ini, but not every gamer can do it. ATI programmers must decide this question.
• NVIDIA: there are cheats as usual...
• Game developers: if you bring into game the anisotropy control, why to hide it from users? Why not to make it available in the options panel!

The subject will be discussed further in the next test because it's based on the same engine.

Unreal II: The Awakening

The speed of the RADEON 9800 is brilliant. NVIDIA has no cheats for this game.

Now let's have a look at the quality level with the trilinear and anisotropic filtering enabled together. At the Performance level the RADEON 9800 has no trilinear filtering at all and the FX 5900 has it very crippled.

RADEON 9800

  

FX 5900

  

Unfortunately, the MIP levels coloring in the Unreal II looks very strange: starting from the second level everything is colored black. However, it excellently shows the border between the MIP levels.

Now we force anisotropy in the Quality mode where the trilinear filtering must be normal:

RADEON 9800

  

FX 5900

  

The trilinear filtering is well seen. Why does it look so complicated in the UT2003 which has benchmarks and why don't we see such problems in the Unreal II that doesn't offer standard tests? Well, business has its own tricks.

RightMark 3D

The victory of the RADEON 9800 is obvious. However, some fans of the FX 5900, especially those who have already bought the 5900 Ultra at very high prices, will keep on asserting that these shaders are not needed in the near future, and when the games become available, all these cards will jump into the middle-end sector. The time will show whether they are right. I understand them: they paid $500 for such a powerful device running at 450 MHz and then saw such a lousy card at 380 MHz which easily beats their monster... Some need speed in the Quake3, some need it in future shader games...