ATI RADEON 9800 PRO 128MB

General information
Peculiarities of ATI RADEON 9800 PRO 128MB
Test system configuration and drivers' settings
Test results: briefly on 2D
RightMark3D synthetic tests: tests philosophy and description
Test results : RightMark3D: Pixel Filling
Test results : RightMark3D: Geometry Processing Speed
Test results : RightMark3D: Hidden Surface Removal
Test results : RightMark3D: Pixel Shading
Test results : RightMark3D: Point Sprites
Test results : 3DMark2001 SE synthetic tests
Summary on the synthetic tests
Test results : 3DMark2001 SE: Game1
Test results : 3DMark2001 SE: Game2
Test results : 3DMark2001 SE: Game3
Test results : 3DMark2001 SE: Game4
Test results : 3DMark03: Game1
Test results : 3DMark03: Game2
Test results : 3DMark03: Game3
Test results : 3DMark03: Game4
Test results : Quake3 ARENA
Test results : Serious Sam: The Second Encounter
Test results : Return to Castle Wolfenstein
Test results : Code Creatures DEMO
Test results : Unreal Tournament 2003 DEMO
Test results : AquaMark
Test results : RightMark 3D
Test results : DOOM III Alpha version
3D Quality
Conclusion

General information

No sooner had we received NVIDIA's GeForce FX 5600 (NV31) and GeForce FX 5200 (NV34) and tested High-End GeForce FX 5800 (NV30), than ATI Technologies attacked again and announced a new-generation line of video accelerators.

As you will see later, the new line comprises the modified and improved products:

RADEON 9800 PRO - 380 MHz chip, 128 MB 340 MHz (DDR 680) 256 bit local memory;
RADEON 9800 - 325? MHz chip, 128 MB 310? MHz (DDR 620?) 256 bit local memory;
RADEON 9600 PRO - 400? MHz chip, 128 MB 300? MHz (DDR 600) 128 bit local memory;
RADEON 9600 - 350? MHz chip, 64/128 MB 275? MHz (DDR 550) 128 bit local memory;
RADEON 9200 PRO - 275 MHz chip, 64/128 MB 275 MHz (DDR 550) 128 bit local memory;
RADEON 9200 - 250 MHz chip, 64/128 MB 200 MHz (DDR 400) 128 bit local memory.

The two latter products are actually copies of the RADEON 9000/PRO with added AGP 8x support. The marketers again gave a higher number of 9200 to the chip weaker than RADEON 9100 (former RADEON 8500LE). Nobody knows the precise specs of 9600 and 9600 PRO, but reportedly, they will be a half of 9800/Pro (4 pixel and 2 vertex pipelines). The rumor has it that 9600/PRO will perform worse than 9500/PRO. When they come out on the market, we will test them thoroughly, but the symptoms are alarming, because NVIDIA's latest mainstream and low-sector products (NV31/34) successfully fight against RADEON 9500/PRO line in the heavy modes (AA and anisotropy).

But time will show. Today we still deal with the High-End solution. By the way, is RADEON 9800 PRO something new regarding technologies and 3D functions, or it is just an overclocked version of RADEON 9700 PRO?

I hope this review will give us the answer. By the way, here is a list of reviews of RADEON 9700/9700 PRO revealing all features and peculiarities of ATI's previous High-End product:

Analysis of RADEON 9700 architecture and Microsoft DirectX 9.0
ATI RADEON 9700 Pro 128MB Review
MAYA II R9700Pro 128MB - performance estimated on the new Pentium 4 2.53 GHz based platform, comparison with the NVIDIA's 40.41 driver
Hercules 3D Prophet 9700 Pro 128MB - new CATALYST 2.3 driver estimated in 3DMark2001 SE, and Unreal Tournament 2003 DEMO final release
PowerColor Evil Commando2 RADEON 9700 Pro 128MB - performance of the new CATALYST 2.3 driver estimated in game tests, 3D quality issues
Sapphire Atlantis RADEON 9700 Pro 128MB - detailed analysis of anisotropic filtering of the RADEON 9700
ATI RADEON 9700, RADEON 9500 64MB and Gigabyte MAYA II RADEON 9500 64MB
Sapphire Atlantis RADEON 9500 128MB - 256bit memory bus of RADEON 9500 128 MB! DOOM III tests
HIS Excalibur RADEON 9700 Pro and tests in the DirectX 9.0 RC0
ATI RADEON 9500 PRO - 128 bit memory bus and buffer compression in AA mode (virtual 256bit bus)
ATI RADEON 9500 64MB, 9500 128MB, 9500 PRO, 9700 and 9700 PRO in DirectX 9.0: Part 1 - Game tests and 3DMark2001, as well as Soft9700!
ATI RADEON 9500 64MB, 9500 128MB, 9500 PRO, 9700 and 9700 PRO in DirectX 9.0: Part 2 - Tests in DirectX 9.0 - synthetic tests from RightMark 3D
YUAN XpertVision RADEON 9000 64MB and RADEON 9700 PRO
Sapphire Atlantis RADEON 9700 & RADEON 9700 PRO Ultimate Edition

RADEON 9800 PRO specification at the time of its announcements:

Fab process: 0.15 micron;
Transistors: 115 M;
Core clock speed: 380 MHz;
Memory bus: 256 bit DDR (DDR II will possibly be supported later);
Local memory size: up to 256 MB;
Memory clock speed: 340 DDR (680) MHz, 24 GB/s bandwidth;
Interface bus: AGP 8x, 2 GB/s bandwidth;
Full support of DX9's main features:

Floating-point 64 and 128 bit data formats for textures (including 3D and cubic textures) and frame buffer (vectors of 4 components of F16 or F32);
Pixel pipelines with floating-point arithmetics (F24[4] or F24[3+1] calculations);
Pixel Shaders 2.0;
4 independent vertex pipelines;
Vertex Shaders 2.0;
N-Patches hardware tessellation with Displacement Mapping, and, if possible, adaptive detail level;
New F-buffer technology supports almost unlimited pixel shaders.

8 independent pixel pipelines
8 texture units (one for pixel pipeline) supporting trilinear filtering without speed loss and a combination of anisotropic and trilinear filtering.
4-channel (4 64-bit channels) memory controller connected to the accelerator's core and AGP switch on the peer-to-peer basis;
HyperZ III+ memory optimization technology (Fast Z Clear and 8x8 depth buffer compression, Hierarchical Z Buffer for fast visibility checking);
Additional optimizations for speedy operation of the double-side stencil buffer.
Early Z test (pixel shaders work only for visible pixels);
Hardware acceleration of MPEG 1/2 unpacking and compression, VIDEOSHADER technology (arbitrary processing of a video flow with pixel shaders);
Two independent CRTC;
Two built-in 10bit 400 MHz RAMDAC with hardware gamma correction;
Integrated TV-Out;
Integrated DVI (TDMS transmitter) interface, up to 2043*1536.
Integrated general-purpose digital interface for external RAMDAC or DVI transmitter and for coupling with TV tuner.
FC packaging (FlipChip).

The characteristics are similar to the previous flagship R300. There are two main differences: optimization of operation with the double-side stencil buffer booting performance in DOOM III engine based games and the like which actively use stencil buffer based dynamic shadows, and special logic in the pixel processors.

This approach makes possible to store in the local memory both final color values calculated by the pixel shader and intermediate values of parameters. It's called F-buffer and requires rather simple chip support (to record and restore the flow of parameters) and a special compiler. This approach allows executing almost unlimited shaders dividing them into parts each of which works independently over a certain screen area and records intermediate parameters for each pixel from this area which are then delivered into the F-buffer for the next shader part. The next shader part reads the parameters in the order they were recorded in, continues calculations and writes down new parameters. This approach is more advantageous than the multipass image rendering. But its support is yet planned only for OpenGL driver.

This technology is meant for realistic graphics and DCC - games are still far from enabling all capabilities of even standard pixel shaders 2.0 limited by 64 computing operations.

In this material the F-buffer technology is closely examined (FIFO-stream buffer).

Card

The card has AGP X2/4/8 interface, 128 MB DDR SDRAM in 8 chips on both PCB sides.

Samsung K4D26323RA-GC2A memory chips, BGA form-factor. Up to 350 (700) MHz; hence the access time of 2.8 ns. By default the memory works at 340 (680) MHz, the chip at 380 MHz.

The memory is the same as on RADEON 9700 PRO.

ATI RADEON 9800 PRO 128MB


ATI RADEON 9700 PRO 128MB

The memory speedup caused the PCB redesigning; the differences from RADEON 9700 PRO are well seen.

The cooler is different as well:

Who suggested the idea of such parallelepiped shape? I think that the idea of a silvery heatsink is bright but such an odd shape... Note that the heatsink is attached to the chip with thermal glue instead of a sticker like in RADEON 9700 PRO. Besides, the heatsink has a small jutter on the bottom.

When we examined RADEON 9700 PRO we noticed that on many cards the die was lower than the protection frame surrounding the chip. And one day we took off the frame so that the heatsink could be better pressed to the die:

RADEON 9800 PRO doesn't have such problem:

The heatsink presses tightly to the die with its jutting part. However, its surface is poorly processed, which is unacceptable for chips with an open die.

The die is of the same size as of R300, but it has a different mount which makes it mirror-like. The package is the same - FCPGA.

The external power connector is replaced with another one, probably because there was a lot of criticism about the fragile connector similar to Floppy Drive. That is why RADEON 9800 PRO houses a more massive socket for a power cable similar to one used for drives (HDD, CDD-ROM).

TV-out doesn't differ from that on RADEON 9700 PRO. See Andrei Karpenko's review about operation of this function.

Overclocking

We'd like to thank Aleksei Nikolaichuk, developer of RivaTuner, for the new version of this utility which provides support for RADEON 9800 PRO:

That is why overclocking is quite possible, and RADEON 9800 PRO demonstrates an excellent potential - 430/385 (770) MHz!

Test system and drivers

Testbeds:

Pentium 4 based computer:

Intel Pentium 4 3066 (HT=ON);
ASUS P4G8X (iE7205) mainboard;
1024 MB DDR SDRAM;
Seagate Barracuda IV 40GB Hard Drive;
Windows XP SP1.

Athlon XP based computer:

AMD Athlon XP 2600+ (2133 MHz);
EPoX (NVIDIA nForce2) mainboard;
1024 MB DDR SDRAM PC3200;
Seagate Barracuda IV 40GB Hard Drive;
Windows XP SP1.

Plus, ViewSonic P810 (21") and ViewSonic P817 (21") monitors.

In the tests we used ATI CATALYST 3.2 drivers (v6.307), VSync off, texture compression off in applications. DirectX 9.0 installed.

Video cards used for comparison:

Reference card NVIDIA GeForce FX 5800 Ultra (500/500 (1000) MHz, 128 MB, driver 42.82 for gaming tests and 42.68 for 3DMark03);
Hercules 3D Prophet 9700 PRO (RADEON 9700 PRO, 325/310 (620) MHz, 128 MB).

Driver settings

The new driver version 6.307 doesn't differ from the previous ones for RADEON 9700. Only SmoothVision and anisotropy versions are set to 2.1 instead of 2. That is why you can look through the settings in the review of RADEON 9700 PRO.

It is possible to force RADEON 9800 in the drivers for RADEON 9700 using Soft9800. This is a patch for RivaTuner which is currently available at NV World; it allows the drivers to detect both RADEON 9700 PRO and RADEON 9800 PRO, and if there are any optimizations only for the latter, it can make them available for RADEON 9700 PRO as well.

We tried to find out what RADEON 9800 PRO offers and what we can get with RADEON 9700 PRO turned into 9800. The results are very interesting. I must say that the rumor stating that

RADEON 9800 PRO = RADEON 9700 PRO + increased clock speed + optimization in the drivers

hasn't proved to be true; the situation is quite complicated in the games.

Also, we tested RADEON 9800 PRO at the frequencies reduced to the level of RADEON 9700 PRO.

Further we will refer to RADEON 9700 PRO working with the patched drivers that define it as RADEON 9800 PRO as to Soft 9800 PRO. If the frequencies are not indicated separately, they are standard for RADEON 9700 PRO - 325/620 MHz

Test results: 3Digest

2D Graphics

Yet Nero, the Roman Emperor, who liked drawing very much was familiar with 2D graphics. At those far times everyone in his circle, even his mother, knew that 2D quality depended on quality of paints (now on a monitor), on brushes (now on a video card), and on artist's view. But that philosophy drove him mad as he considered 2D an ideal to be worshiped, and he even killed his mother... But Nero's spirit hasn't calmed down yet and it often appears like a horror in front of those who got blurry images... Only if you have reliable shielded cables and hgih-quality monitors Nero's spirit will never find his shelter there...

That is why we 2D tests were carried out with ViewSonic P817-E monitor and Bargo BNC cable. RADEON 9800 PRO has perfect 2D quality! It perfectly works at 1600x1200@85Hz and 1280x1024@120Hz.

But remember that quality depends also on a certain sample.

RightMark 3D synthetic tests (DirectX 9)

Today we will describe and run the suite of synthetic tests we are currently developing for the API DX9.

The developed RightMark3D test suite now includes the following synthetic tests:

Pixel Filling Test;
Geometry Processing Speed Test;
Hidden Surface Removal Test;
Pixel Shader Test;
Point Sprites Test.

The philosophy of these synthetic tests and their description are given in NV30 Review.

Those who are eager to try RightMark 3D synthetic tests can download the "command-line" test versions which record the final XLS file in the XML format accepted in Microsoft Office XP:

Every archive contains description of test parameters and an example of a .bat file used for benchmarking accelerators. We welcome your comments and ideas as well as information on errors or improper behavior of the tests.

Mailto: unclesam@ixbt.com.

Practical Tests

Below are the data obtained with budget and mainstream accelerators based on two major families (ATI and NVIDIA).

ATI:

RADEON 9700 PRO 128 Mb
RADEON 9800 PRO 128 Mb

NVIDIA:

GeForce FX 5800 Ultra 128MB

Pixel Filling

As the latest Catalyst drivers (3.1 and higher) show rather dubious results in our synthetic test on some ATI's cards (probably because of the failed attempt to optimize this test in the modes with texture sampling), the results of Pixel Filling with textures for RADEON won't be published, only the case without textures:

Product	Theoretical maximum, M texel/s.	Measured maximum, M texel/s.
RADEON 9700 PRO	2600	2335
RADEON 9800 PRO	3000	2546
GeForce FX 5800 Ultra	2000	1995

NVIDIA is closer to its theoretical peak though ATI is not far either; only RADEON 9700 PRO could have it better if it were not for the limiting memory bandwidth.

Geometry Processing Speed

Fixed TCL performance (performance of the shader that emulates it):

In case of the fixed TCL the performance gap corresponds to the difference in the frequencies.

Vertex Shaders 1.1:

it's the same with Vertex Shaders 1.1

Shaders 2.0 with loops:

Well, ATI has optimized performance of the second shaders and, in particular, loops; the performance gain is sometimes double! I think it depends on the drivers rather than on the hardware section, and I hope RADEON 9700 will also get such boost with the future drivers. But now these optimizations are included only into R350 to demonstrate, as I think, the advantages of R350 over R300 for those who do not go into details.

But it is also possible that low performance of the second shaders was caused by the hardware and it could already be corrected; soon we will find out it.

So, with the new drivers R350 looks better in this test than NV30. The second shaders are now as efficient as the first ones, and we can say that loops are granted for free. The driver optimization is well noticeable (in particular, for the compiler and shader microcode optimizer) - almost double growth. I wonder if NVIDIA is going to surprise us soon like it often happened before?

Cross dependence on geometry detail level and shader's complexity:

The performance balance remains the same irregardless of scene complexity - it seems that the size and algorithms of vertex queues and caches were not changed.

Hidden Surface Removal

HSR efficiency in different resolutions in a complex scene (without and with textures):

the slight difference is caused by the side effects - the difference is the frequencies. Any change of the HSR algorithm or its efficiency would make the scores more different.

HSR efficiency depending on scene complexity (without and with textures):

The single-level HSR of NVIDIA prefers simple scenes while ATI's hierarchical algorithm performs best of all in average scenes, but textures bring closer these approaches regarding the behavior and effectiveness. Ideally, NVIDIA's HSR is less efficient but in reality textures and early Z culling make this difference inconsiderable.

Pixel Shading

Have a look at the performance of pixel shader s in R350 in comparison with R300:

Point Sprites

(lighting off/on):

The gap is a bit greater than the difference in the clock speeds because of the optimized second vertex shaders. Now, in contrast to R300, NV30 fails in all modes.

3D graphics, 3DMark2001 SE synthetic tests

All measurements in 3D are taken at 32bit color depth.

Fillrate

Multitexturing:

R350 and R300 take positions according to the difference in the clock speeds. In the multitexturing mode NV30 takes the lead thanks to 8 texture units at the higher frequency of 500 MHz, but it loses in the single-texture mode because of twice fewer pixel pipelines.

Pixel Shader

Simple variant:

The scores get much better in the higher resolutions. But as we found out earlier (in RightMark 3D), performance of the pixel shaders changes according to the frequency growth.

Probably, it's because of another optimization of ATI's new drivers for R350 working this time only with pixel shaders and in this particular test.

Let's see whether pixel shaders with more intensive calculations make an effect:

Here the performance corresponds to the frequencies, i.e. these marks can be considered adequate.

Vertex Shaders

The results are actually expected.

In general, 3D Mark 2001 scores match RightMark 3D, though they carry less information because synthetic tests parameters are unadjustable.

Summary on synthetic tests

Let's sum up the detailed examination of various units of R350 in the synthetic tests.

The architectural differences are quite unimportant. You can regard R350 as R300 with the increased core speed, enhanced optimization of operation of the double-side stencil buffer (which will tell upon in DOOM III based games) and potential hardware support of F-biffer and long pixel shaders.
Remember that ATI's drivers enable certain functions only for R350; it might make you think that the architecture of R350 is much superior to R300. But if the new drivers providing the same gain for R300 are released, the difference in the test scores will depend only on that in the clock speeds, which is rather small. Even if the developers never let them out, there may be third-party modifications and patches (for example, in RivaTuner) making these optimizations work on R300. Well, let's see how the events will unfold.

3D graphics, 3DMark2001 game tests

Anisotropy was set to 16x for ATI's cards and to 8x for NVIDIA because algorithms of this function considerably differ (we discussed it in NV30 Review). The criterion is just one: maximum quality. The screenshots were shown several times already. Besides, it's interesting to compare NVIDIA's different anisotropy modes (Application, Balanced, Aggressive) with the ATI's high-quality mode; our readers can estimate how speed and quality correlate looking at the screenshots from NV30 Review demonstrating anisotropic quality.

The tables give us all necessary data, that is why I won't comment them separately, except Soft9800.

3DMark2001, 3DMARKS

Note that forcing of RADEON 9800 with AA and anisotropy enabled on RADEON 9700 PRO boosted up the performance relative RADEON 9700 PRO with AA and anisotropy. Is it because the drivers are selectively optimized? Sure! Though there is almost no any gain without AA and anisotropy.

3DMark2001, Game1 Low details

3DMark2001, Game2 Low details

3DMark2001, Game3 Low details

3DMark2001, Game4

Let's sum up the results of 3DMark2001SE:

Without the extra load the RADEON 9800 PRO is not much far ahead of GeForce FX 5800 Ultra (sometimes it even loses). In the heavy modes it can badly lose (Game4) or grasp the lead (up to 40-50% in Game3, Game2). At least, we know for sure that thanks to the 256-bit bus and higher memory throughput, RADEON 9800 PRO looks perfect in the AA mode.
The drivers are optimized for RADEON 9800 PRO judging by the comparison of Soft9800 PRO and RADEON 9700 PRO in the heavy modes with AA and anisotropy. But without this load the former still nudges out the latter. Note that at the increased clock speed and with AA and anisotropy enabled, Soft9800 PRO is almost equal to RADEON 9800 PRO.

3D graphics, 3DMark03 game tests

3DMark03, 3DMARKS

We can witness the same situation with optimization of ATI's drivers for RADEON 9800 PRO, when replacing Device ID of 9700 PRO with that one of 9800 PRO makes the performance goes up.

3DMark03, Game1

Wings of Fury Test characteristics:

DirectX 7.0; approx. 32000 polygons on the scene, 16 MB memory used for textures, 6 MB for buffers for vertices and 1 MB for indices.
All geometrical operations are based on Vertex Shaders 1.1 which can be emulated with CPU (if there is no hardware support).
All planes have 4 texture layers, that is why accelerators able to process 4 textures in a pass will benefit.
Fire and tail effects are made with the point sprite and other techniques.

3DMark03, Game2

Battle of Proxycon:

DirectX 8.1; Approx. 250 000 polygons on the scene with Pixel Shaders 1.1 (and 150 000 polygons on the scene with Shaders 1.4), 80 MB memory used for textures, 6 MB for buffers for vertices and 1 MB for indices.
All geometrical operations are based on Vertex Shaders 1.1 which can be emulated with CPU (if there is no hardware support).
All heroes are "dressed" with the vertex shaders as well.
Some light sources made dynamic shadows with a stencil buffer.
All pixel operations are carried out with shaders 1.1, and if possible with shaders 1.4.
Calculation of per-pixel lighting for haze effects and other components.
Accelerators supporting pixel shaders 1.1 use one pass for determining Z buffer, then 3 passes for each light source. If an accelerator supports shaders 1.4, it needs one pass for each light source.

3DMark03, Game3

Trolls' Lair:

DirectX 8.1; approx. 560 000 polygons on the scene with Pixel Shaders 1.1 (and 280 000 polygons on the scene with Shaders 1.4), 64 MB memory used for textures, 19MB for buffers for vertices and 2 MB for indices.
All geometrical operations are based on Vertex Shaders 1.1 which can be emulated via CPU (if there is no hardware support).
All heroes are "dressed" with vertex shaders as well.
Some light sources made dynamic shadows with a stencil buffer.
All pixel operations are carried out with shaders 1.1, and if possible with shaders 1.4.
Calculation of per-pixel lighting for haze effects and other components.
Realism of the heroine's hair is achieved with physical models and anisotropic lighting.

3DMark03, Game4

Mother Nature:

DirectX 9.0; approx. 780 000 polygons on the scene, 50 MB memory used for textures, 54MB for buffers for vertices and 9 MB for indices.
Every leaf is separately animated with Vertex Shaders 2.0. Grass is animated with vertex shaders 1.1.
Lake's surface is formed with pixel shaders 2.0.
Sky is made with pixel shaders 2.0, sun glints are formed with extra-precision calculations in DX9.
Earth surface is made with shaders 1.4.

Summary on 3DMark03?

NVIDIA proved with its v42.68, that it's not a problem to make optimization for this packet. RADEON 9800 PRO outscores NV30 only in and loses in the others.
Optimization of ATI's drivers for RADEON 9800 PRO is well seen in this case as well.

3D graphics, game tests

3D games used to estimate 3D performance:

Return to Castle Wolfenstein (MultiPlayer) (id Software/Activision) - OpenGL, multitexturing, Checkpoint-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, Grand Cathedral demo, test settings: quality, S3TC OFF

Quake3 Arena v.1.17 (id Software/Activision) - OpenGL, multitexturing, Quaver, 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 Demo (Digital Extreme/Epic Games) - Direct3D, Vertex Shaders, Hardware T&L, Dot3, cube texturing, default quality

Code Creatures Benchmark Pro (CodeCult) test demonstrates operation of cards in the DirectX 8.1, Shaders, HW T&L.

AquaMark (Massive Development) test demonstrates operation of cards in the DirectX 8.1, Shaders, HW T&L.

RightMark 3D v.0.4 (one of the game scenes) - DirectX 8.1, Dot3, cube texturing, shadow buffers, vertex and pixel shaders (1.1, 1.4).

Quake3 Arena, Quaver

NVIDIA's optimization of the drivers for this game makes RADEON 9800 PRO lose to GeForce FX 5800 Ultra.

But ATI's drivers are well optimized to RADEON 9800 PRO, which helps Soft9800 PRO outpace RADEON 9700 PRO by 25% at the equal clock speeds.

Strangely, Soft9800 PRO clocked at the frequency of 9800 PRO runs faster than RADEON 9800 PRO (with AA and anisotropy). They have overdone it :-). This fact means that RADEON 9800 PRO is not simply a copy of 9700 working at the higher clock speeds.

Serious Sam: The Second Encounter, Grand Cathedral

ATI likes this test more than NVIDIA :-). Although RADEON 9800 PRO loses to its rival in the anisotropy mode, it wins in the heavier one with AA and anisotropy enabled.

Are the drivers of RADEON 9800 PRO optimized for this game? Sure! But it can be seen only for AA and anisotropy modes. Again, Soft 9800 PRO performs better at 380/680 MHz than RADEON 9800 PRO, that is why the affirmation that RADEON 9800 PRO = RADEON 9700 PRO + increased frequencies is inconsistent.

Return to Castle Wolfenstein (Multiplayer), Checkpoint

Again NVIDIA's optimization for this game? Strangely enough but RADEON 9800 PRO loses the battle again even in the heavy modes with AA and anisotropy.

The drivers are optimized for 9800 but you can feel it only when AA and anisotropy are used.

This test confirms that RADEON 9800 PRO is a product different from RADEON 9700 PRO.

Code Creatures

In this test the shader speed is very important. ATI's cards do not have any problems here, in contrast to NVIDIA GeForce FX 5800. That is why RADEON 9800 PRO wins.

Well, if a test is not widely used, it won't be optimized for RADEON 9800 PRO. In this test the latter takes the lead exactly at the expense of its higher clock speeds and the GPU optimizations (caches, a higher shader speed etc.).

Unreal Tournament 2003 DEMO

The vertex shader speed has a strong effect on this game, and RADEON 9800 PRO goes far ahead. In the heavy modes with AA and anisotropy the new-comer looks brilliant!

Optimization of the drivers for 9800 is pretty good, as we can see it at the frequencies of 9700 PRO. 9700 PRO overclocked up to 9800 PRO and turned into Soft9800 PRO showed the scores in AA and anisotropy modes equal to RADEON 9800 PRO, that is why we can speak about pure optimization.

AquaMark

Without the extra load RADEON 9800 PRO grasps the victory (except 1600x1200), but with AA and anisotropy it loses to its competitor, though it works speedier in the AA mode.

RADEON 9800 PRO has its drivers optimized for this test, but the test is not that crucial and the developers didn't do their best :-)

RightMark 3D

On one hand, RADEON 9800 PRO smashes NVIDIA's card because the test much depends on the shaders speed; on the other hand, this test is not open for the programmers at ATI or NVIDIA, and we can see that Soft 9800 PRO 380/680 MHz falls far behind from RADEON 9800 PRO in the heavy modes! Moreover, at 325/620 MHz it lags even behind RADEON 9700 PRO!

When it comes to games or tests for which the developers can't make any optimizations, when the processor works by itself and when a given test actively uses modern technologies, RADEON 9800 PRO shows that its not simply an improved version of 9700 PRO.

DOOM III Alpha

NVIDIA's drivers, as well as the more flexibly programmable NV30 help this company win.

3D Graphics Quality

We have carefully examined performance of RADEON 9800 PRO in games, in particular with AA and anisotropy enabled, and found no difference in quality from RADEON 9700 PRO. In one of our previous reviews we looked into the differences between NVIDIA and ATI regarding ATI and anisotropy, you can read about it in GeForce FX 5800 Ultra Review (theory, practice), and in GeForce FX 5600/5200 Review (practical comparison of the anisotropy).

Conclusion

Today we were dealing with the fastest game graphics accelerator - ATI RADEON 9800 PRO. So, what do we have to say about it?

The launch of this product is ATI's response to NVIDIA's announcement of NVIDIA GeForce FX 5800 Ultra, though a bit ephemeral. The Canadian company did it best to squeeze maximum of the 0.15 micron process, and first of all, higher clock rates. However, in some tests RADEON 9800 PRO has lost in spite of the driver release. As you know, GeForce FX hasn't got the driver release yet and it's not clear how much it will boost the performance. Well, we expected the weightier gain of ATI, at least, no defeat. ATI might bring out more efficient modification of R350, for example, under the name of RADEON 9900 PRO; besides, we do not exclude that it can couple its modification with DDR2 memory.
RADEON 9800 PRO is not just a chip with the increased frequencies. It's obvious that R350 was redesigned regarding a higher output of valid chips, frequencies and interaction of the units and caches (i.e. performance in applications). It's hard to separate the benefit from architecture optimization and from driver optimization, but we will be able to do it if drivers providing optimization for R300 are released.
The developers are cheating. The drivers are optimized for a number of popular tests so that RADEON 9800 PRO looked a faster runner than RADEON 9700 PRO. The efforts of our programmer Aleksei Nikolaichuk can help owners of RADEON 9700/PRO speed up its operation using the Soft9800 patch.
Judging by the layout the video card has almost the same cost as RADEON 9700 PRO, and the final price will be influenced by the cost of R350 and by the marketing interests. I think it won't be dearer than $400-420 in the beginning.
It doesn't make sense for owners of RADEON 9700 PRO to sell their cards to buy 9800 PRO. But those who want a top High-End accelerator and who are thinking about 9700 PRO had better wait for RADEON 9800 PRO. Reportedly, such cards will get to the shelves not earlier than the second half of April. The accelerated version of GeForce FX 5800 codenamed NV35 should appear on the market at the same time. That is why it's not clear which models RADEON 9800 PRO is going to stand against: NV30 or NV35. What is going to happen with RADEON 9700 PRO? I wish 9700 and 9700 PRO remained on the market but stepped down the price ladder ($200-250); but as you know, new RADEON 9600(PRO) are positioned for $170-200, and speedier RADEON 9700 might spoil its sales. Besides, R300 is not much cheaper than R350, and it's hardly beneficial for ATI to cut prices for R300. I suppose RADEON 9700/PRO will entirely be replaced with RADEON 9800/PRO.

Andrey Vorobiev (anvakams@ixbt.com)

Alexander Medvedev (unclesam@ixbt.com)