EliteGroup AG200

EliteGroup Computer Systems (ECS) is one of the biggest manufacturers of entry-level mainboards. Lately, the company also started graphics cards production. ECS decided on graphics chips from SiS: SiS315, Xabre400 and Xabre200, a Xabre600-based graphics card is also coming soon. EliteGroup has always targeted the value sector where the price is the crucial factor. That's why the company chose SiS, not ATI or NVIDIA. The chips from SiS were aimed exactly at this market sector. We don't want to say that the Xabre core is a poor quality solution. Never! First of all, it was the first graphics chip to support AGP 8x. This chip also supports DirectX8.1, that is, can process pixel and vertex shaders.

ECS AG200 graphics card comes in a nice retail box. Besides the card itself, the package includes a CD with the drivers and a user's manual. No cables for connection to a TV-set are provided.

SiS Xabre200 chip installed in ECS AG200 works with the SiS301 chip, which implements TV-Out support. Depending on the version of SiS301, it can have an integrated transmitter for image output through the DVI port onto LCD displays, a second RAMDAC for dual-display configurations, a DAC for output onto a TV-set and a special stereo glasses support unit.

AG200 also features a SiS301MV chip with TV-Out support. EliteGroup was one of the few manufacturers who used the ability of SiS301MV to be a synchronizer for stereo glasses. That's why AG200 is a unique Xabre-based product with integrated output onto semiconductor liquid crystal shutter (LCS) glasses.

The graphics card bracket carries (from left to right in the snapshot): a stereo glasses port, TV-Out (Composite), TV-Out (S-Video) and D-Sub Out leading to analog display. Actually, SiS301MV supports image output through a digital DVI port, but it's not put into life in the AG200 graphics card.

The card has 64MB DDR SDRAM memory in eight chips from Samsung with 5ns clock cycle time.

The graphics memory works at its nominal frequency of 200MHz (400MHz DDR). It heats up very little at work and doesn't require heatsinks.

Overclocking

The working frequencies of the card are 200MHz/200MHz (400MHz DDR). Actually, SiS recommends setting the frequencies to 200MHz for the chip and 166MHz (333MHz DDR) for the memory for Xabre200 based cards, so the graphics memory in the ECS AG200 is initially set at a higher rate. During overclocking, we could make ECS AG200 work at 245MHz chip and 245MHz (490MHz DDR) memory. The chip and memory frequencies growth made 45MHz or 22.5%. Not bad, and we even think that this graphics card could have reached even better rates if we had used some additional cooling.

SUMA Platinum K2

Imagination Technologies introduced the STM KYRO II graphics chip in 2001. Actually, KYRO II is an overclocked version of the older KYRO. The first KYRO worked at 125MHz, while KYRO II works at 175MHz. There are few graphics card makers who had KYRO/KYRO II based products in their graphics product range. Today, these cards are a rare bird. The one we have got was produced by the Korean SUMA Company.

In general, KYRO II was an interesting chip and KYRO II based graphics card stands out in our roundup. First of all, it's the only card based on a chip that has been discontinued by now. So, KYRO II based cards can vanish from the market at any moment. Secondly, the KYRO chips were the only mass-production solutions with tile rendering (more on tile rendering can be found in our STM Kyro II Review: Hercules 3D Prophet 4500 64MB Graphics Card). Moreover, SUMA Platinum K2 is the only graphics card in our roundup to have SDRAM memory (not DDR). KYRO II graphics chip is obsolete now, but we still put it at the same level with RADEON 8500LE and GeForce3 Ti200 because of its relatively high price.

SUMA Platinum K2 card comes in a transparent box. Besides the card, the box contains a user's manual, a CD with drivers and an S-Video-to-Composite adapter.

Nearly all KYRO II based cards followed the reference design developed by Videologic. SUMA Platinum K2 graphics card is no exception. The PCB is covered with black lacquer; the small cooler is glued to the chip surface and is hard to remove.

First KYRO II based cards had 32MB of memory onboard, then the 64MB variant appeared. Our SUMA Platinum K2 has exactly 64MB SDRAM in eight chips soldered up to both sides of the PCB.

The graphics memory from EliteMT has 5ns clock cycle time and the nominal working frequency of 200MHz. As we have said, KYRO II uses ordinary SDRAM memory, not DDR.

The box of SUMA Platinum K2 reads "K2 64MB SE". We might think that the card is based on KYRO II SE graphics chip and many price-lists mention it as KYRO II SE. But the nominal frequencies of KYRO II SE are 200MHz chip and 200MHz memory, while SUMA Platinum K2 works at 175MHz/175MHz respectively. So, we deal with an ordinary KYRO II, no "SE" here.

Overclocking

Judging from our experience with KYRO and KYRO II based cards, we are not going to see any great overclocking results here. Due to its architecture, KYRO II can only be overclocked by 5-10MHz. This chip works synchronously with the memory, i.e. at the same frequency, and the chip and memory frequencies should be increased simultaneously during overclocking.

We were not mistaken in our suppositions. The maximum we managed to squeeze from the card were 185MHz/185MHz. If we could increase the memory frequency independently, it could reach 200MHz, which is the nominal frequency of 5ns SDRAM. We will see the overclocked KYRO II performance in benchmarks later on. For now, let's look at the table that lists the characteristics of all the graphics cards participating in our roundup:

Let's see whether the cards are worth their price. The benchmarks are waiting ahead.

Testbed and Methods

The point of our today's review is to test the performance of contemporary value ($50-100 price range) graphics cards in modern games. The testbed was configured as follows:

• Intel Pentium 4 2000MHz CPU;
• ABIT IT7-MAX2 (i845E) mainboard;
• 256MB PC2100 DDR SDRAM;
• IBM DTLA 15GB 7,200rpm HDD;
• 24x CD-ROM;
• Windows XP Professional.

The following games and benchmarks were used:

• Quake3 Arena 1.30;
• Unreal Tournament 2003 Demo;
• WarCraft III;
• AquaMark 2.0.

The graphics cards worked with the following drivers:

• NVIDIA Detonator 40.72 WHQL
• ATI CATALYST 2.5
• SiS Xminator II ver.3.08 WHQL
• STM KYRO 1.06.16.00.51

The drivers' settings were left default, the only exception was the disabled vertical synchronization (V-Sync OFF). Xminator II driver for SiS Xabre based graphics cards allows adjusting the texturing quality in OpenGL and Direct3D applications. Depending on the settings, the rendering quality by Xabre can be made higher at the expense of speed, or vice versa. We tested ECS AG200 card with default settings, where texturing quality is average. We also set 32-bit color depth and enabled maximum graphics quality settings in all the games. As we deal with value cards, we decided to carry out our tests in low resolutions: 800x600, 1024x768 and 1280x1024. Every graphics card was tested at its nominal working frequencies and after overclocking.

Performance

Quake3 Arena

Notwithstanding its considerable age, Quake3 Arena is still popular among fans of first person shooters. This game uses OpenGL API and is not very pressing on the hardware by today's criteria. For our benchmarks, we set all the game settings at maximum, turned on 32-bit textures without compression and used tri-linear filtering. Then we ran Four.dm_66.dm3 demo in Timedemo mode.

Below are the screenshots we made for every graphics card tested, so that you could compare the picture quality provided by each of them:

Inno3D Tornado3 Ti200	ABIT Siluro GF4 MX	SAPPHIRE Radeon 8500LE
SAPPHIRE Radeon 9000	ECS AG200	SUMA Platinum K2

It's clearly seen that Xabre200 based ECS AG200 graphics card offers the worst image quality. KYRO II-based SUMA Platinum K2 is a little better. NVIDIA GeForce3 Ti200 and GeForce4 MX440 have nearly the same quality, while RADEON 8500LE and 9000 based cards from SAPPHIRE provide the best picture in this game.

Quality checked, let's compare the speeds of the cards.

As we see, SAPPHIRE RADEON 8500LE is the fastest in Quake3 Arena in the both modes: overclocked and nominal. Inno3D Tornado3 Ti200 and ABIT Siluro GF4 MX are close behind. The graphics cards based on Xabre 200 and KYRO II chips are the very last ones in this race.

Unreal Tournament 2003

Unreal Tournament 2003 was released in the end of 2002. It's a good alternative to Quake3 Arena for benchmarking various computer subsystems. The game works in Direct3D and features integrated means to measure graphics card performance. We set all the graphics quality settings to "High" and ran Flyby test in DM-Atlantis map. The numbers come later on, right now let's check the image quality.

Inno3D Tornado3 Ti200	ABIT Siluro GF4 MX	SAPPHIRE Radeon 8500LE
SAPPHIRE Radeon 9000	ECS AG200	SUMA Platinum K2

And again, Xabre 200 based EliteGroup AG200 provides the worst image quality: you can clearly see the texture MIP levels borders on the floor. All other cards, including SUMA Platinum K2, were surprisingly good at rendering the scene. There is no definite difference between them, so let's go over to the numbers.

We see a larger gap in performance between the graphics cards here. SAPPHIRE RADEON 8500LE is again outperforming the others. Inno3D Tornado3 Ti200 is a little behind. The rest of the cards are 2-3 times slower than the leaders. Curiously, Xabre 200 based ECS AG200 shows similar results in all resolutions. This suggests that the drivers for Xabre are not optimized yet for the card to work at its full capacity.

Note also that ABIT Siluro GF4 MX based on GeForce4 MX440 GPU fell far behind Inno3D Tornado3 Ti200. It's all because Unreal Tournament 2003 uses vertex shaders. Xabre 200 and GeForce4 MX440 graphics chips execute them on the software level, while KYRO II doesn't support them at all. That's why there is such a big gap between the cards based on these chips and the cards using the chips, which were developed as high-end solutions at first: GeForce3 Ti200, RADEON 8500LE and even RADEON 9000 all feature hardware vertex shaders support. One more reason why GeForce4 MX440 performed slowly is that Unreal Tournament 2003 requires laying up to three textures per pixel, while GeForce4 MX440 can only render two textures per clock and has to waste one more clock to lay the third texture.

WarCraft III

Graphics cards are usually tested in first person view games or simulators where the life of the protagonist may depend on the speed of the graphics subsystem. But lately there have appeared a number of 3D games that belong to a different genre: real-time strategies, role games and others. A striking example of such a gaming application is WarCraft III. This game actively uses the 3D graphics cards features and pleases the eye with its beautiful graphics. Unfortunately, WarCraft III has no built-in means of measuring the graphics card performance, so we had to use a specific methodology to be able to test in this application.

Together with the game, we ran FRAPS ver.1.9 program that can measure the fps rate in any Direct3D application. This program can also show average fps value for a specified period of time. The user has to manually determine the start and end measurement points by pressing the [Scroll Lock] key. WarCraft III has no demos to use for testing purposes, but there are a lot of scripts in the game that are based on the game engine. So, we can use one of these scripts as a demo. The script we used was "Chasing Visions" from the first episode of the "Exodus of the Horde" campaign. The measurements were made from the very beginning until the user takes the action.

There are two points to bear in mind: FRAPS uses CPU resources too, and thus affects the overall result. But its impact is rather low and, moreover, all the graphics cards were under the same conditions. So, we can claim that the results are reliable. The second point is somewhat subtler. The start and end points of statistics gathering were marked by our pressing the [Scroll Lock] key. The overall result may slightly vary depending on the exact moments when the key was pressed. After long-time preparations and training, we put forth a volunteer who had the highest Dexterity stat of all of us and could start and stop the measurement process at the "black screen" moment: before the movie started and just after it finished. So, we can claim that the results are more or less valid, although we wouldn't disregard some measurement error, of course.

The graphics quality in RTS games like WarCraft is even more important than in FPS games. Let's see what our cards can boast here:

Inno3D Tornado3 Ti200	ABIT Siluro GF4 MX	SAPPHIRE Radeon 8500LE
SAPPHIRE Radeon 9000	ECS AG200	SUMA Platinum K2

Curiously enough, SUMA Platinum K2 graphics card provides the best graphics quality here. Look at the trees leaves and grass patches: they look lusher when drawn by KYRO II. The same greenery looks less impressive by Inno3D Tornado3 Ti200 and ABIT Siluro GF4 MX. SAPPHIRE RADEON 8500LE and RADEON 9000 graphics cards render the scene less picturesque than KYRO II, while Xabre 200 based ECS AG200 makes the textures look quite fuzzy which is clearly seen from the stones and closer trees. Let's now check the performance of the cards:

SAPPHIRE RADEON 8500LE is again ahead of all. KYRO II is the slowest here, although provides best graphics quality. The performance results of other cards look quite natural. The only thing to mention is that Xabre 200 outperformed GeForce4 MX440. But if we recall the not very impressive image quality provided by this chip in the "default" mode, the performance will no longer be that important here.

AquaMark 2.0

AquaMark 2.0 uses the graphics engine from AquaNox game, which is a mixture of arcade, shooter and futuristic submarines simulator. It is a rather old game, but it has excellent graphics. The AquaMark benchmark uses a rather dark scene, so it's hard to evaluate the perfection of the graphics quality in it. That's why we decided to show no screenshots. AquaMark 2.0 also uses Direct3D. It's possible to specify the amount of textures to be used in the demo. As we deal with 64MB graphics cards, we chose 24MB of textures, which is more than enough for modern games.

SAPPHIRE RADEON 8500LE is once again showing its superiority. The only mode it loses to the overclocked Inno3D Tornado3 Titanium200 is 800x600. ABIT Siluro GF4 MX is the last but one here and loses even to Xabre 200 based ECS AG200. SUMA Platinum K2 is a clear outsider.

Conclusion

The price difference between the most expensive and the cheapest product reviewed is around $50. Someone would think it's considerable; others may call it a trifle. But what do we get for the extra money? Let's compare the cards, starting from the most expensive one.

SAPPHIRE RADEON 8500LE and Inno3D GeForce4 Ti200 differ by only $3 in price. But the first one features DVI output onto an LCD display and, moreover, it's the unrivalled leader of our today's test session both: in speed and picture quality in 3D applications. These two graphics cards used to belong to high-end products, but now their price is quite affordable. They both support pixel and vertex shaders, have well debugged drivers and reveal no problems at work.

The next echelon consists of ABIT Siluro GF4 MX, ECS AG200 and SAPPHIRE RADEON 9000. The latter is the best here. This graphics card outperforms the other two in all benchmarks, except Quake3 Arena, where it loses to ABIT Siluro GF4 MX. It has its shortcomings, though: no DVI output for displays connected via the digital port and no overclocking opportunities. In all other respects, it's a worthy competitor to GeForce4 MX440 based graphics cards.

ECS AG200 deserves a separate word. We are accustomed to graphics cards based on chips from ATI and NVIDIA, while the graphics solutions from SiS are still a novelty. This graphics card has an appealing price, and the graphics core itself supports AGP 8x, pixel shaders and DirectX8. The performance of Xabre 200 is comparable with that of GeForce4 MX440: Xabre 200 manages to win in two of the four benchmarks. But still, the low texturing quality and raw drivers don't add anything positive to this baby. Its price should be even lower, so that it could really compete in this market segment.

To our regret, KYRO II chip seems quite out-dated now. This solution could compete with GeForce3 or RADEON 8500 a year ago, when there were no fast Pentium 4 and Athlon XP processors. Driver updates for KYRO II come out rarely and we might never see their new version as the developer of this graphics chip, Imagination Technologies, gave up its graphics division. To tell the truth it doesn't make much sense to buy KYRO II now: a lot of PC games now use T&L and other technologies, which are not available by KYRO II. Besides, the price for SUMA Platinum K2 seems too high without any convincing reasons.

Soon, after the release of GeForceFX, the price for GeForce4 Titanium 4200 based graphics cards may drop down below the $100 mark. ATI may also reduce prices of its RADEON 9500 PRO / 9500 chips. Xabre 400/600 based cards will be their immediate rivals. Then we will have a new situation in the value graphics market. Let's not make any forecasts now and just wait until the price reductions, so that we could invite you to check one more benchmarking spree.