Sapphire Atlantis RADEON 9500 128MB and Video Cards Tests in the DOOM III v.0.02

General information
Features of the video cards
Test system configuration and drivers' settings
Test results: briefly on 2D
Test results : Fillrate synthetic test from 3DMark2001 SE
Test results : 3DMark2001 SE gaming tests
Test results : Quake3 ARENA
Test results : Serious Sam: The Second Encounter
Test results : Return to Castle Wolfenstein
Test results : Comanche4 DEMO
Test results : Unreal Tournament 2003 DEMO
Test results : RightMark 3D
Test results : DOOM III v.0.02
Conclusion

Not so long ago we reviewed a new ATI's product released for the price niche of $130-200. That was RADEON 9500. In that article we discussed why it was so important to have video cards on the middle level - NVIDIA has almost nobody to compete against in this sector. The RADEON 9000/Pro is a competitor against the entry-level cards from NVIDIA (GeForce4 MX); the RADEON 8500 which is a bit stronger than the RADEON 9000 Pro is all by itself (it has a higher price than the GeForce4 MX has, but it's inferior to the GeForce4 Ti 4200 in performance. In fact, it takes an intermediate position).

The whole price sector of $150 to $220 (which is the most popular among gamers) is totally controlled by NVIDIA and its partners (it includes GeForce4 Ti 4200 of all kinds). That is why it was a thoughtful decision of ATI to bring out a new line able to stand against the above mentioned NVIDIA's solutions. Unfortunately, SIS is not able to create a card with qualitative and quantitative parameters close to the Ti 4200, not to mention other companies. Matrox left the market long time ago; although its Parhelia is close in speed to the Ti 4200, it's pricier than the RADEON 9700 Pro. The PowerVR has unclear prospects at all. That is why ATI is the only firm that can be competitive here.

Here is the RADEON 9500/9700 family:

RADEON 9700 PRO - 325 MHz chip (8 pipelines), 128 MB 310 MHz (DDR 620) 256bit local memory;

RADEON 9700 - 275 MHz chip (8 pipelines), 128 MB 270 MHz (DDR 540) 256bit local memory;

RADEON 9500 PRO - 275 MHz chip (8 pipelines), 64/128 MB 270 MHz (DDR 540) 128bit local memory;

RADEON 9500 - 275 MHz chip (4 pipelines), 64 MB 270 MHz (DDR 540) 128bit local memory.

This official list will probably have to be corrected, but later on it. And now, before we go straight to the analyses of the RADEON 9500, let's have a look at the reviews dealing with the new line of ATI.

Theoretical materials and reviews of video cards which concern functional properties of the VPU ATI RADEON 9700 (Pro) / RADEON 9500 (Pro)

Analyses of architecture of the RADEON 9700 and Microsoft DirectX 9.0
ATI RADEON 9700 Pro 128MB review
MAYA II R9700Pro 128MB - performance estimated on the new testbed based only on the Pentium 4 2.53 GHz, comparison with the NVIDIA's 40.41 driver
Hercules 3D Prophet 9700 Pro 128MB - performance of the new CATALYST 2.3 driver estimated in 3DMark2001 SE, and Unreal Tournament 2003 DEMO final release
Sapphire Atlantis RADEON 9700 Pro 128MB - detailed analyses of anisotropic filtering of the RADEON 9700
ATI RADEON 9700, RADEON 9500 64MB and Gigabyte MAYA II RADEON 9500 64MB

In the RADEON 9500 review we found out that this product doesn't have a performance level suitable for its price of $140-150. That is why its defeat to the Ti 4200 based card brings to naught all advantages of its DX9 support in many cases. Besides, the RADEON 9500 loses to its rival even in the modes with enabled AA and anisotropy. That is why such cards shouldn't be priced over $120.

Why do we return to the RADEON 9500 repeating the base review rather than estimating the production card? There are two reasons: either the today's card is based on a new, cheaper, design, or the influence of the 128MB cards is so great that it has to be studied thoroughly.

Below you will see that the design is exactly the same. But the fact that the card comes with 128 MB memory instead of 64MB boosts the performance much.

The RADEON 9700 Pro was released so much time ago but the 9500 model hasn't yet received a separate PCB or a separate chip. Why? Why did ATI decide to cut the memory size of a RADEON 9700 based card and cut off the chip? Cards based on a 8-layer PCB with a 256-bit bus are quite expensive (as we mentioned in the review on the RADEON 9500). Besides, it uses the same expensive R300.

The analyses of certain information from reliable sources shows that the cost price of such PCBs is not so high anymore, because the production technique is already mastered both by PC Partner (which manufactured all RADEON 9700 Pro in the beginning) and by the partners most of which have already started making such cards and PCBs. As we know, a product goes to the mass market when the expenses on its production become as low as possible. As for the chips... The same sources notify that the production of the VPU R300 is getting stable, that is why the prime cost of them is falling down as the percentage of valid dies grows up.

Therefore, having estimated the benefits of production of separate PCBs and chips, ATI and its partners chose the approach of "Velocity 100", i.e. they assemble the RADEON 9500 on the same 9700 model with the memory chip "shortened". It's possible that they use graphics processors that failed to work at the frequencies specified for the RADEON 9700 Pro, because the 9500 is to be clocked at 275 MHz against 325 MHz of the RADEON 9700 Pro. Though it's also possible that they strip off normal chips. However, even in this case there are a lot of questions to be answered: for example, is such cut-off implemented on the software level or on the hardware one? We'll get back to this issue later.

What's the key difference between the 64MB RADEON 9500 and its brother with 128MB memory? Let's take a look at the card.

Card

Sapphire Atlantis RADEON 9500 128MB 64 MB

Sapphire Atlantis RADEON 9500 128MB 64 MB
The card has AGP x4/x8 interface, 128 MB DDR SDRAM located in 8 chips on both PCB sides. The memory chips are from Hynix marked as HY5DU283222 F-36, BGA form-factor. The highest frequency is 275 (550) MHz. The memory runs at 270 (540) MHz, the chip at 275 MHz.

Comparison with the reference design, front view
Sapphire Atlantis RADEON 9500 128MB	Reference card ATI RADEON 9700 Pro 128MB

ATI RADEON 9500 64MB

Comparison with the reference design, back view
Sapphire Atlantis RADEON 9500 128MB	Reference card ATI RADEON 9700 Pro 128MB

ATI RADEON 9500 64MB

NO differences from the PCB of the RADEON 9700 Pro! Note that each chip of the above mentioned marking has a 32bit access. That is why having mounted all 8 chips the manufacturer leaves the 256bit bus as well!

The RADEON 9700 64MB card was left with 128 bits when they took off 4 chips - taking into account that each of four remaining chips was 32bit, the total capacity amounted to 128 bits.

And this card comes with the 256-bit interface! That is why the readers should remember this surprise from ATI:

Until they develop a design specially for the RADEON 9500 based cards (and it's probable that they will never do it, because production of cards based on the current design becomes cheaper with time), we should remember that the 64MB and 128MB cards differ only in the memory exchange interface: 128 and 256 bits. This can't help having an effect on the performance, though the VPU is stripped off markedly, and even a 256-bit bus would hardly help the 4-pipeline chip change its speed considerably (but our tests will show that the speed gain is noticeable).

When we studied the RADEON 9500 for the first time we noticed that the chip is marked as the RADEON 9700 Pro

So, what do we have?

RADEON 9500 (just for today?) is built on the RADEON 9700 Pro, having no different features from it except the memory chips with a higher access time of 3.6ns;
64MB model has twice as less memory chips, that is why the memory exchange bus becomes 128bit;
128MB model is a copy of the RADEON 9700 Pro, they have no difference outwardly (even without the coolers);
The chip of the RADEON 9500 is the same, and I don't see how it's possible to cut it on the hardware level without correcting the marking of the processor (to make it clear which chip is crippled and which is not).

In view of this I think that the RADEON 9500 is cut off on the software level. If you remember, in the 3dfx Velocity100 they locked one module on the software level to make its speed slower compared to the Voodoo3 2000. But that time they didn't provide any reliable protection, the key for the second TMU was easy to find.

The programmers from ATI are smarter. They tried to protect a lot of cards from overclocking, in particular, RADEON 9000 - whatever frequency you set the speed doesn't change. The RADEON 9700 has a cunning BIOS: its frequencies often remain unrecognized by tweakers.

That is why if we actually have the RADEON 9700 instead of RADEON 9500, it can be a real problem to remove the software protection and turn the latter into the former. However, this aspect is yet to be studied.

Now look at the cooling system.

Sapphire Atlantis RADEON 9500 128MB
To make the card different from the reference sample the developers from Sapphire replaced the cooler. We underlined it yet in the review of the Sapphire's line. The cooler is similar to those used on the GeForce4 Ti 4200 cards - hot air is blown out of the center toward the heatsink's ribs.

I'm going to make no more comments on the design of the card as we have already reviewed a lot of various RADEON 9700 Pro based solutions and covered a wide range of aspects of such PCBs.

Overclocking

When I took off the heatsink I saw that the die is almost as high as the protection frame around the chip which the heatsink is pressed against, that is why the chip is cooled effectively (we used Arctic Silver thermo grease).

And such high effectiveness couldn't help boosting the overclocking potential.

Sapphire Atlantis RADEON 9500 128MB

275/540 -> 380/660 MHz

Excellent, isn't it? :)

Note

in course of overclocking you must provide additional cooling, in particular, for the card (first of all, for its memory):

overclocking depends on a definite sample, and you shouldn't generalize the results of one card to all video cards of this mark or series. The overclocking results are not the obligatory characteristics of a video card.

Test system and drivers

Testbed:

Pentium 4 based computer (Socket 478):

Intel Pentium 4 2530 MHz;
ASUS P4T533 (i850E) mainboard;
512 MB 32bit RDRAM PC4200;
Seagate Barracuda IV 40GB;
Windows XP.

The test system was coupled with ViewSonic P810 (21") and ViewSonic P817 (21") monitors.

In the tests we used ATI's drivers 6.200. VSync was off in the drivers, texture compression was off in the applications. DirectX 8.1 was used. Texture detail level set to High Quality.

For comparison we used results of the following video cards:

Leadtek WinFast A170T (GeForce4 MX 440, 270/200 (400) MHz, 64 MB, driver 40.72);
MSI GF4MX460 (GeForce4 MX 460, 300/275 (550) MHz, 64 MB, driver 40.72);
Triplex Millennium Silver GeForce4 Ti 4200 (250/256 (512) MHz, 64 MB, driver 40.72);
Albatron Medusa GeForce4 Ti 4200 (250/222 (444) MHz, 128 MB, driver 40.72);
Gigabyte MAYA RADEON 8500 (275/275 (550) MHz, 128 MB);
ATI RADEON 8500 (275/275 (550) MHz, 64 MB);
Hercules 3D Prophet 9000 Pro (RADEON 9000 Pro, 275/275 (550) MHz, 64 MB);
Hercules 3D Prophet 9000 Pro (RADEON 9000 Pro, 275/275 (550) MHz, 128 MB).

Drivers' settings

The DirectX 9 drivers are already available, but they are now going through the beta tests, that is why operation of the RADEON 9500\9700 in the DX9 will be estimated next time.

All the settings were discussed in the reviews listed above.

Test results

2D graphics

Let's start with 2D. The 2D quality is really perfect, like in case of the previous RADEON 9700 Pro based cards!

Remember that estimation of 2D quality can't be objective. It depends on quality of a sample, and a card/monitor tandem (you should pay special attention to quality of the monitor and the cable).

3D graphics, 3DMark2001 SE - Fillrate

All measurements in the 3D tests were carried out at 32-bit color depth.

Fillrate

In this test we are to prove that the RADEON 9500 128MB has a 256-bit bus indeed. Look at the first scores. We enabled the AA2x mode (the lightest AA) knowing that it didn't cause speed drops in the RADEON 9700Pro. Taking into account that the RADEON 9500 is deprived of half of the pipelines, it won't be easy for it to cope with the AA even with a 256-bit bus, that is why we simplified the task as in this test the chip doesn't have much work to do. In 1024x768 the speed remains the same, but after that it goes down.

In the multitexturing mode when the chip may use all its possibilities in operation with the textures, the performance doesn't worsen at all when the mode changes for AA2x! This fact proves that the card comes with a 256-bit bus!

3D graphics, 3DMark2001 - gaming tests

3DMark2001, 3DMARKS

3DMark2001, Game1 Low details

Test type	Gap between RADEON 9500 128MB and RADEON 9500 64MB, %
1280x1024 without AA and ANISO	5,7
1600x1200 without AA and ANISO	11.7
Test type	Gap between RADEON 9500 128MB and GeForce4 Ti 4200 128MB, %
1280x1024 without AA and ANISO	-20.8
1024x768 with AA4x	19.3
1024x768 with ANISO 8x/16xQ	-26.9

3DMark2001, Game2 Low details

Test type	Gap between RADEON 9500 128MB and RADEON 9500 64MB, %
1280x1024 without AA and ANISO	14.0
1600x1200 without AA and ANISO	16.7
Test type	Gap between RADEON 9500 128MB and GeForce4 Ti 4200 128MB, %
1280x1024 without AA and ANISO	0.3
1024x768 with AA4x	70.2
1024x768 with ANISO 8x/16xQ	34.6

3DMark2001, Game3 Low details

Test type	Gap between RADEON 9500 128MB and RADEON 9500 64MB, %
1280x1024 without AA and ANISO	13.5
1600x1200 without AA and ANISO	15.9
Test type	Gap between RADEON 9500 128MB and GeForce4 Ti 4200 128MB, %
1280x1024 without AA and ANISO	-6.2
1024x768 with AA4x	74.9
1024x768 with ANISO 8x/16xQ	48.9

3DMark2001, Game4

Test type	Gap between RADEON 9500 128MB and RADEON 9500 64MB, %
1280x1024 without AA and ANISO	18.2
1600x1200 without AA and ANISO	20.4
Test type	Gap between RADEON 9500 128MB and GeForce4 Ti 4200 128MB, %
1280x1024 without AA and ANISO	6.4
1024x768 with AA4x	38.1
1024x768 with ANISO 8x/16xQ	-37.3

It's obvious that the 256-bit bus helps a lot to such a weak and crippled chip. At least, in the AA modes the RADEON 9500 outstrips the Ti 4200. We haven't tested yet in the combined (AA+anisotropy) mode, but I reckon the ATI's solution won't yield to the GeForce4 Ti 4200 either.

3D graphics, gaming tests

For estimation of 3D performance in games we used the following tests:

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, all settings - maximum: detailing level - High, texture detailing level - #4, S3TC OFF, smoothness of curved surfaces is greatly increased with variables r_subdivisions "1" è r_lodCurveError "30000" (at default r_lodCurveError is 250!), the configurations can be downloaded from here

Comanche4 Benchmark Demo (NovaLogic) - Direct3D, Shaders, Hardware T&L, Dot3, cube texturing, highest quality

Unreal Tournament 2003 Demo v.1027 (Digital Extreme/Epic Games) - Direct3D, Vertex Shaders, Hardware T&L, Dot3, cube texturing, default quality, ANTALUS

RightMark Video Analyzer v.0.4 (Philip Gerasimov) - DirectX 8.1, Dot3, cube texturing, shadow buffers, vertex and pixel shaders (1.1, 1.4).

Quake3 Arena, Quaver

Test type	Gap between RADEON 9500 128MB and RADEON 9500 64MB, %
1280x1024 without AA and ANISO	13.8
1600x1200 without AA and ANISO	18.2
Test type	Gap between RADEON 9500 128MB and GeForce4 Ti 4200 128MB, %
1280x1024 without AA and ANISO	-20.3
1024x768 with AA4x	8.8
1024x768 with ANISO 8x/16xQ	-18.2

Serious Sam: The Second Encounter, Grand Cathedral

Test type	Gap between RADEON 9500 128MB and RADEON 9500 64MB, %
1280x1024 without AA and ANISO	11.2
1600x1200 without AA and ANISO	15
Test type	Gap between RADEON 9500 128MB and GeForce4 Ti 4200 128MB, %
1280x1024 without AA and ANISO	34
1024x768 with AA4x	92.9
1024x768 with ANISO 8x/16xQ	18.3

Return to Castle Wolfenstein (Multiplayer), Checkpoint

Test type	Gap between RADEON 9500 128MB and RADEON 9500 64MB, %
1280x1024 without AA and ANISO	25.8
1600x1200 without AA and ANISO	26.8
Test type	Gap between RADEON 9500 128MB and GeForce4 Ti 4200 128MB, %
1280x1024 without AA and ANISO	3.2
1024x768 with AA4x	35.7
1024x768 with ANISO 8x/16xQ	4.4

Comanche4 DEMO

Test type	Gap between RADEON 9500 128MB and RADEON 9500 64MB, %
1280x1024 without AA and ANISO	14
1600x1200 without AA and ANISO	156.7
Test type	Gap between RADEON 9500 128MB and GeForce4 Ti 4200 128MB, %
1280x1024 without AA and ANISO	2.7
1024x768 with AA4x	43.6
1024x768 with ANISO 8x/16xQ	-13.9

Unreal Tournament 2003 DEMO

Test type	Gap between RADEON 9500 128MB and RADEON 9500 64MB, %
1280x1024 without AA and ANISO	5
1600x1200 without AA and ANISO	6.7
Test type	Gap between RADEON 9500 128MB and GeForce4 Ti 4200 128MB, %
1280x1024 without AA and ANISO	-2
1024x768 with AA4x	29.8
1024x768 with ANISO 8x/16xQ	2.6

RightMark 3D

Test type	Gap between RADEON 9500 128MB and RADEON 9500 64MB, %
1280x1024 without AA and ANISO	8.2
1600x1200 without AA and ANISO	9.7
Test type	Gap between RADEON 9500 128MB and GeForce4 Ti 4200 128MB, %
1280x1024 without AA and ANISO	32.3
1024x768 with AA4x	43.8
1024x768 with ANISO 8x/16xQ	0.7

DOOM III v.0.02

We decided to test various video cards in the alpha version of the DOOM III which recently leaked into the Net. Although this "half-game" comes with its own demo program, we used two programs provided by 3DNews.ru. They are demo002 and demo006. To find them go to the DOOM III v.0.02 review published on this site.

Well, there is not much that can be said about this game. Remember though that the SIS Xabre switched off the light on the demo006, and the player was hovering around in the dark. No credits. The Matrox Parhelia showed some abstract pictures where it was impossible to understand where a player was and what was happening. No credits either. The KYRO II failed to start up in this game at all.

Conclusion

The main conclusion is that the RADEON 9500 128MB is much faster than its 64MB sibling; it's a direct competitor against the GeForce4 Ti 4200, in the heavy modes it even outshines the latter despite ts crippled graphics processor. Remember also that it does support the DirectX 9.0, in contrast to the Ti 4200.
In the beginning that card will be priced at $150-160, which seems to be a good price tag for such card.
It's probable that the RADEON 9700 was turned into the RADEON 9500 on the software level. But it's only a supposition, we don't promise that with the 9500 you will get the 9700 free. The examination is not completed yet as there is one more interesting aspect to cover - the chip is coupled with the 256-bit bus, though it was expected that the 9500 would have both the pipelines and the memory controller stripped down (in this case the chip wouldn't be able to send more than 128 bits at a clock cycle). It's possible that such reduction was made exactly on the software level, and it's different in case of the 64MB and 128MB cards, while the chip is the same.
The Sapphire's video card is an example of excellent quality, stability and reliability.
As for the new division of the RADEON 9500 64/128MB line, ATI will probably take measures to change the respective specifications not to mislead users, because at the moment the RADEON 9500 is bound to the 128-bit bus, and originally there was no word concerning a 128MB card. The RADEON 9500 Pro is getting more mysterious as the developers are to leave 128MB memory on the PCB from the RADEON 9700 and cut the bus off to 128 bits.

Andrey Vorobiev (anvakams@ixbt.com)

Write a comment below. No registration needed!