Gigabyte Maya II R9500 Video Card Review

By: Dmitry Zinovyev

There is no and can't be the ideal 'price to quality' relation for an offered goods. That is, if you want to buy a good graphics accelerator cheap, you won't get it. It's always possible to get a poor goods at high price though. This is reality where it's sometimes very difficult to find the right balance between the price and quality. There is no concept of good and bad here, they are all replaced with lots of notions, and if you go deep down them you'll either get mixed up completely or spend an awful lot of time which could have been used for better advantage and pleasure. Many see the future where the whole gaming industry as well as the industry of professional graphics are moving in the ability to make a scene as realistic as possible, so thousands of man-years and immense engineering efforts of largest leading chip development companies are being spent. That's why Microsoft released the new API - DirectX9, with ATI and Nvidia backing it up with their R300 and NV30 chips. While NV30 chips have not yet seen the retail shops, ATI has been selling its offsprings for quite a while to all those after on-screen realism.

Every company has its own flagship product that attracts the public by its performance and potentials. And if we join the arguments between the two companies as to whose product is better, it makes sense comparing just the flagship products of theirs. All the remaining squadron remains in the shade. Today we'll be looking into a product that has no analogs, and not because there aren't video cards more productive than that, but because there are no cards offering this 'price-quality-functionality' ratio. Note that I don't mean to say this ratio is the best, - I mean it has no analogs. This is the ATI 9500 Pro.

As is known, the ATI R300 chip underlying the 9500 and 9700 video cards is in a way revolutionary, and its revolutionary character is just about the ability to make use of cutting-edge technologies improving the visualization quality, e.g. programmable vertex and pixel shaders etc., but the major trait that makes the new generation of accelerators stand out is the full-featured support for API DirectX 9. The flagship product ATI 9700 Pro is still rather costly for today to persuade the people who buy new computers or upgrade the existing choose it for their needs. Very few can afford to fork around $350+ for this video card - that take up almost one third or even half the price of the whole computer. Anyway, many want to get a video card that won't restrict the user functionally for at least the coming one and a half year, even though the card is going to be a performance overkill. So, to get a better idea of the differences between the 9500 and 9700 series, look at the below table.

As is seen from this table, today we're testing a card the weakest of the whole ATI's R300 line. Theoretically, we can expect its performance to be a bit as twice as low compared to the top-end ATI9700 Pro card and on par with 8500 cards. For details of the potentials demonstrated by video cards based on the R300 chip, read the materials on our web-site. But we are now moving on to the particulars.

Gigabyte Maya II R9500

As always, let's first run through the package bundle. The card comes shipped in a standard box bearing the Maya logo and is colored to dark tints.

In the part of packaging, Gigabyte has gone slightly against the established fashion and released an easy to open box. Gigabyte does not supply this card wrapped in an inner box the way Abit or Asus do. For a card positioned as the cheapest of the line, the package bundle appears to be quite good. It includes:

• A DVI -> D-Sub adapter for plugging in a second monitor.
• A S-Video + Composite cable for plugging in a TV-set.
• A drivers CD.
• PowerDVD xp 4.0 (DVD player).
• Serious Sam (Game).
• Rune (Game).
• Oni (Game).
• 4x4 EVO (Game).

The card itself is made in ATI's smart PCB design. It's amusing to see that ATI's 9500 and 9700 cards have practically identical PCB design and absolutely identical GPUs of the same production line. While the 128-bit bus is organized at the expense of reduced memory chips, the disconnection of 4 pipelines inside the chip is evidently organized logically. By now there are ways to enable all the 8 pipelines, although it's not yet possible to get one-one results, but we'll take about it in what follows. Outwardly, the card is absolutely identical to ATI 9700Pro, anyway there is a difference although completely invisible to the eye. This becomes vivid upon a bit closer look at both cards. Changes in the 9500 card are seen in the output modules as well as in some added modules close to the AGP connector.

The connector for additional power merits a special consideration. ATI engineers decided the card wouldn't do without additional power, so a connector similar to that of the FDD was positioned on the farthest corner of the card. As I mentioned earlier, the card comes bundled with a splitter cable for power connectors, so there shouldn't be any problems with plugging it in. As was already mentioned, there is a standard R300 chip onboard manufactured on the 36th week of the year 2002, which is about early September.

The 64 MB memory chips made by Hyundai-Hynix are installed on both sides of the PCB and have the BGA form factor. As I already said, the quantity of chips has been reduced to 4, with their standard remained intact, which resulted in the drop of bit-rate from 256 to 128 bit. Judging by the marking HY5DU283222 F-36, the memory features a 3.6 ns access time.

DVI, TV-Out, D-Sub

As you can see, there are three connectors on the header of the video card. One DVI, one S-Video for plugging in a TV-set, and one D-Sub 15. Of priority is the D-Sub 15 connector, with the DVI and S-Video sharing the second RAMDAC of the chip. That is, simultaneous use of the TV-Out and a device connected to the DVI is impossible. The quality of TV-Out is middling since with the default settings the screen is crispy, and with the increased contrast many parts of bright half-tints turn bleached. The splitter cable for connecting the TV is organized in an awkward way. It's just the cable, not the adapter, which is too short and no extension cords are supplied in the bundle. It's been the first time I had to carry a VCR with a TV set to the computer in order to be able to test the output while testing the video card. If we plug only a TV set to the video card with all the monitors unplugged, then the loading and the text mode will be run in the NTSC standard, which is rather poor for our region and there are no ways currently to change the situation.

Cooling system

The radiator on board the Gigabyte card differs from the Reference in only the design and color, while its dimensions and cooling quality seem to be the same as before. Like in the whole like of R300-based video cards, the memory has no cooling of its own, but nothing prevents you to organize it yourself. The only remark as to the installation of third-party devices for memory cooling is the presence of electric cells amid the chips, so coupled radiators won't do - only individual cooling of each separate chip matters. But on the whole, if you don't intend to overclock, memory cooling is not going to be much of a destabilizing factor, unlike the chip cooling. The thermopaste between the radiator and the chip leaves much to be desired. It seems like a hybrid of double-sided scotch tape and paper is becoming trendy among video card manufacturers, but this fashion came up not because of the efficiency of this cooling method but due to its cost savings. So we recommend replacing it with a more decent option like KPT-8 or the like thermopaste. It's also worth reminding that some R300 chips have a manufacturing flaw in the skirting standing above the main crystal. We wrote about this issue in our previous review devoted to 9700Pro, so we're not going to dwell on that.

Some unpleasant remarks.
It is mandatory to note that the bug with crispy screen seems to be widespread and is really annoying for the user. I noticed that problem still on the Sapphire 9700Pro, but that time I thought it was an issue of the particular video card specimen, - moreover, I sent that card out for guaranteed replacement. On this Gigabyte's video card the bug has been reproduced, and what is more, I received many letters from users who suffer from similar problems. There were revealed several versions of this problem. In one of the versions, the stripes are always visible regardless of which outlet the output device (TV, monitor) is connected to or which drivers are installed. In the other version, the stripes are seen only when a monitor is plugged to the D-Sub connector, - all the remaining outputs produced normal image. In the third version, version two was eliminated through replacing Catalyst 3.0 drivers with Catalyst 2.5 drivers. My video card specimen had this version two problem, and replacement of drivers didn't help. I also revealed some dependence on the make of motherboards. With Intel-850EMV P4(i850Å), the problem was highly visible, the same with Gigabyte 7VAX (KT400), but with Asus A7V8X (KT400) the stripes were not seen at all. There was no dependence on the power supply and the AGP settings, so this issue is still open.

Another small problem was found out thereafter. Not all yoghurts are equally good to your health regardless of what TV commercials say - that's how I formulated the problem as soon as it came to my head. The thing is that not all monitors are recognized by the DVI output. If you wish to connect the master monitor to the DVI-out via an adapter, to get rid of the mentioned waves on the screen, your monitor can cancel displaying the text mode, i.e. the loading of the operating system with all the BIOS settings. Of course, as soon as the loading of the video driver is complete the image will resume, but before that the monitor won't switch on. This bug definitely depends on the monitor. Two absolutely identical Intergraph 21" monitors having different connections (one made via the D-Sub, the other via the BNC) showed contrary results - one worked, the other wouldn't. An old 14" monitor also wouldn't run. This is all I have been able to try by now. That parameter does not depend on the DVI -> D-Sub adapter, so it may easily turn out that even a monitor having only a DVI-connection will refuse to run correctly on this output. By the way, I tested operation of the monitors via adapters with a GeForce4 used. All the monitors without exception do run correctly.

It seems like ATI's Catalyst dubbed by folks as "cataclysm" is anyway not free of the bugs and blunders despite ATI's insistent assurance of the contrary. I wonder how long it will take the user to learn how to switch between the 60Hz and 85 Hz or higher of screen refresh rate under Windows XP. As for me, it took about ten minutes before it finally came upon me that it should have been changed in THREE places, whereas the monitor does switch but in fact the former 60 Hz stays. The first two places are the tabs "Adapter" and "Monitor" in Windows XP, with the third being the "Monitor" tab in the ATI control panel. This looks like a blunder in the drivers... but bugs merit a special consideration. I am now bringing in only those which I remember better, but I am sure they were much more than that. The most annoying bug was related to the RC0 drivers update management under DirectX 9. Do you really like the look-and-feel of this control panel?

In the end, whatever I did I was unable to change the screen refresh rate in a monitor plugged to the D-Sub ... there was simply no third place for changes. Then, after disabling the V-Sync in the drivers control panel (seems like I didn't do anything else) the OpenGL driver happened to disable for an unknown reason so badly that I had to re-install the drivers. Another thing happened while I was playing in Morrowind when half the screen was covered with white noise of green dots and then the computer hung. The same bug also reproduced while playing Doom3 without the system hang. I can't remember the driver versions with which all that happened (at least, no less than Catalyst 2.5), but the very fact already suggests some ideas.

Overclocking

This is a large topic as related to video cards based on the chip, since that kind of overclocking was effected not only about frequencies but also about unlocking the pipelines within R300. As I mentioned earlier, most ATI's video cards based on R300 are made on practically identical design, except for the ATI 9500 Pro whose PCB design differs from the generic, but this time we are not talking about this card. Since our card is an ATI 9500, has 64 MB memory onboard, we won't be able to push the bit-rate up to 256 bit because of the PCB organization and chip structure, where the 128 bit is arranged due to reduction of memory chips from 8 to 4; there's wiring for the remaining chips on board, but they are simply not welded. The first PCB remake of 9500 into 9700 implied hardware intrusion, or in simpler terms, using a soldering iron. We won't consider this option since it is quite faulty, e.g. loss of guarantee servicing upon unsuccessful attempt. The second option is about software-driven remake. The idea behind it is that some lines of the assembly code inside the drivers are changed, which results in that a 9500 chip becomes detected as a 9700 and the system therefore turns on using it to the full. I must express my deep gratitude to Unwinder who in his RivaTuner utility created and included the Soft9700 patch-script which turns the risky recasting R9500 video cards into an easy and revocable task. Unfortunately, far not all R9500 video cards are able easily turning to 9700 from 9500. This is related to that the R300 assembly line does not yield 100% fully functional chips, - screen-out specimens come into producing cheaper models of the card. This resembles production of CPUs where the processor is marked by the frequency it is able running at. This seems to be a similar case, - on R9500 cards they sometimes mount R300 chips having either faulty HyperZ unit, not used in original R9500 cards, or unusable pipelines. In the case of a faulty HyperZ, you will anyway be able running the video card albeit with the "chessboard" on the background, but the card won't start and hang upon loading the driver if the pipelines are unusable. You will easily see if the HyperZ module is faulty already in the 2D by the chessboard cells displayed around all the icons.

By the way, if there arises an incredible situation when the bugs are seen in the 2D only while all is fine in the 3D, simply reduce the "hardware acceleration" property in "Monitor" -> "Diagnostics" by two units. Currently, the statistics of overclocks shows approximately 70% success versus 30% failures. Unfortunately, for now there hasn't been a technology that allows disabling the unusable HyperZ module without affecting the pipelines. Nevertheless, we can't rule out such functionality and in the nearest future the RivaTuner utility will feature an undo-able software-driven recasting of 9700 cards into 9500 :) as well as software-driven recasting of amateur 9700 cards into professional FireGL cards.

It's a pity out tested card has fallen just within those 30% of failed overclocks. The unusable HyperZ module appeared to be quite a nice-looking bug because all the games started in the overclocked mode looked as if viewed through Robocop's eye at his most fierce times ... occasional ripples, jittering letters and all cells around the screen... Nevertheless, there is also one good old type of frequency overclocking left, which showed the following results. I was able setting the chip to the boundarily stable 385 MHz frequency, with all the other frequencies above that limit being easily set, but upon starting any 3D application all simply fell off to the operating system. The memory started running stably at 337 MHz. But we'll dwell on that a bit later on.

Benchmarks

To test the performance of the Gigabyte R9500 video card based on the R300 chip, we assembled these two testing systems:

Testing system No. 1 (Hi-End):

• Processor Intel Pentium 4 2.8 GHz;
• Motherboard - Intel 850EMV (i850E, 512 Mb RDRAM PC800), AGP 4x);
• Sound card Creative Sound Blaster Audigy;
• HDD Seagate Barracuda ATA IV 7200 rpm.

Testing system No 2 (Mid-Range):

• AMD Athlon XP2000+ processor;
• Motherboard Asus A7V8X (KT400, 256 MB DDR PC2700 memory, AGP 8x);
• Creative Sound Blaster Live;
• HDD - IBM IC35AVVA07 7200 rpm.

We tested the performance in the following benchmarks and applications:

• Return to Castle Wolfenstein  - Checkpoint demo (OpenGL, Quake 3 core);
• Unreal Tournament 2003 Demo (Direct3D stress test benchmark);
• DooM III E3 Demo (extremely resource-hungry overkill benchmark with shaders, bump-mapping, OpenGL and lots of other stuff to be topical by the end of 2003);
• Codecreatures demo (complex DirectX 8 benchmark);
• 3DMark 2001SE (synthetic Direct3D benchmark).

The only remark as to the installation of third-party devices for memory cooling is the presence of electric cells amid the chips, so coupled radiators won't do - only individual cooling of each separate chip matters. We are trying to give an all-out estimation of the card performance when used in various applications on different platforms.

3DMark 2001 SE

This benchmark is the most impartial since the accelerator is tested throughout the whole range of 3D applications, and the results are made up of data received from an aggregate of various sub-benchmarks. Unfortunately, you can't test the cutting-edge features of the new accelerators with this benchmark, but this is not yet topical since other accelerators don't support them either yet, and to date there are still no games made especially for DirectX9 and ATi Radeon 9700 features. Those cool new benchmarks standing on par with these won't appear earlier than new state-of-the-art accelerators matching these functionalities hit the market. We are expecting the release of Unreal 2 and 3DMark 2003 especially for testing the performance of both state-of-the-art chips and Microsoft's new API DirectX 9.

I must admit though, using the 128-bit memory bus instead of 256-bit with 4 pixel pipelines disabled has not affected the performance positively at all. Frankly, although that was expected originally, such a performance drop is slightly shocking indeed. At low resolutions, the performance stays in the middle between GeForce 4 Ti4200 and GeForce 3, which is not bad at all. There was no processor dependence detected since the graph exhibits an absolutely linear behavior. But starting from the 1280x1024 resolutions, the situation has changed sharply and at the highest resolutions the R9500 card demonstrated a complete failure. This can be explained by two merely two things, with the first being the presence of 64 MB memory onboard, which is certainly not enough. Another thing is that the AGP operation is either poorly optimized or inadequate. To understand that, let's look at the following graph, since the second testing system demonstrates full functioning of the AGP8X.

As you can see from the graph, the general situation with performance is much better than in the previous testing system. In the low resolutions, the performance of the card is closer to GeForce 4 Ti4200 and this trend remains preserved while diminishing completely upon reaching the top. The first idea that comes into my head is that in this situation it was using the AGP8x that played the decisive part. It's also possible that ATI drivers are optimized for technologies used on AMD platforms *3DNow)... We'll clarify this aspect in our further reviews.

Well, I've rarely seen such graphs. It is evident that using the R9500 on the AMD platform with support for AGP8x results in an essential performance !increase! in higher resolutions. The gap at higher resolutions is in fact equal to the reverse gap at the low resolutions.

Return to Castle Wolfenstein

This benchmark can be freely regarded as a processor dependence test. This benchmark is most frequently used for testing weak video cards since its requirements to the graphics accelerator can't be regarded as high enough, because the Quake3 engine is quite an easy task for modern accelerators.

Once again, the 9500 performed poorly and stayed on its former position. I wish I saw identical results up to 1280x1024 resolution, but no miracle occurred and the card demonstrated a result a bit similar to that in the previous 3DMark benchmark except the drop at higher resolutions. That is no wonder, - there shouldn't have been any drop since this game uses rather small textures as per today's standards which easily fit within the 64 MB of local memory.

But on that platform at low resolutions the processor power proved insufficient anyway, and the result is practically identical up to 1600x1200. At 1600x1200, the result becomes similar to that for GeForce 4 MX400. It's worth noting that the only result higher than MX400 was produced at 1280x1024 since low resolutions strongly depend on the processor, while the higher depend much on the card itself. Judging by the produced results, we can assume that higher resolutions are not a strong part of R9500 cards.

In this benchmark, as I already mentioned, it was the processor that played the leading part, which is confirmed by the graph. If there were much more geometry and textures, the result would have been in fact identical. So if you are up to playing old games, the powerful processor is a must.

Unreal Tournament 2003 Demo

This benchmark is completely reverse to the previous. All they have in common is that they are all in the group of gaming tests. This benchmark exhibits powerful geometry in the frame and large textures, is absolutely not processor-dependent, and all depends on the card itself.

As always, the low resolution is practically identical in all of the video cards (which is really encouraging), so the outsiders, GeForceMX 400 and SiS Xabre 600, are immediately visible. In this benchmark the tested card stays on par with GeForce 3, slightly taking a lead over it at 800x600 and a bit lagging behind it at 1600x1200. Just want to note that the performance drop as compared to the flagship 9700Pro is quite substantial, and at high resolutions the difference becomes greater than two times.

And again the R9500 in this benchmark has performed quite well. The performance of R9500 is undoubtedly higher than that of GeForce MX and is slightly lower than GeForce Ti 4200. The ratio of performance levels does resemble that on the first sting system, and hence we arrive at some interim conclusion. Unlike 3DMark, there is practically no difference made by using AGP8x.

The graph here is self-evident, which says if you intend to play Unreal Tournament at resolutions higher than 1024x768 it doesn't matter what processor you are using, of course within reasonable limits. At low resolutions, there does exist processor dependence, but that is a regularity. Nevertheless, Unreal Tournament is a game, and requirements to the hardware resources have got to be reasonable, therefore it is unable loading a modern accelerator to the full.

Codecult Codecreatures demo

This benchmark is a real torture for the video subsystem but it shows no processor dependence, since the processor is not a bottleneck. It outputs up to 500000 polygons per frame and makes use of the vertex and pixel shaders of version 1.1, that is why it does not run on Nvidia MX-Series video cards.

In this benchmark, like in many others, the operation speed of ATI R9500 is on par with GeForce 3 and much lower than the higher-end card based on the identical R300 core, - ATI R9700Pro. It means that in complex game scenes which are likely to come across quite often in new games there will be a lack of speed even with the hardware support for DirectX9.

On this graph, the situation is very similar to the previous. But the gap between the 64 MB GeForce 4 Ti4200 and the card in question has turned much narrower at 1600x1200.

There is practically no processor dependence, that is, using this card with a relatively weak processor in games of ultra-heavy graphics won't demonstrate any essential performance difference.

DooM III E3 Demo

And lastly, here comes a gaming benchmark of super-heavy graphics :). This is in fact a benchmark, not a game; and it can be called a test with a great reserve, since the final version will demonstrate considerable performance differences on specific types of video cards thanks to the effected optimization. Nevertheless, we believe the situation will change radically, so we are using this test version as a full-featured performance benchmark. Reminding it again that we are not using the second testing platform to verify the performance in this benchmark, since 256 MB of RAM is required to produce correct results.

At this benchmark, the ATI's card performed quite well. On the whole, the situation is very similar to Unreal Tournament 2003, but at low resolutions it's a long way to go to the processor dependence. The ratio of speed characteristics is practically identical in all the resolutions, however, it's worth mentioning that the tested card demonstrated a considerable advance over GeForce 3 at low resolutions which then became matched as the resolutions were going up, while at 1600x1200 the loss proved minimized.

Overclock

Regular frequency overclocking produced a pretty good performance boost and approached the level of GeForce 4 Ti4200-4600. It was found out during the tests that the chip speed has less effect on the performance as compared to the memory speed. There was an impression that the chip wouldn't overclock at all - only the figures changed on the screen. Only when we reduced the values to the extreme minimum we were able to find out that the performance did change, but after setting the maximum speed to 415 MHz all hung up (which was really expected). As was already mentioned, 384 MHz on the chip proved to be stable, which is simply fantastic frequency, almost 100 MHz higher than the pre-set values. Provided the R300 chip is used reasonably, the level of overclocking is likely to be lower, since more units will be activated and thus more heat will be emitted. As for the memory, the operating frequency 337 MHz is also not bad for this type of memory, and much depends on its frequency. This means the main restricting factor is the bus cut down to 128 bit.

Findings

If you are a fancier of exotic hardware, this card will appeal to you... it's hard to call it middling. On the one hand, it offers full-featured functionality which will be in demand in the nearest future, plus a change of recasting the card into a slightly cut-down version of the 9700 card. On the other hand, it features low performance, still buggy drivers, unstable operation in some applications, plus a high chance of getting a complete bunch of ATI's hardware bugs like stripes stretching over the screen or elevated skirting on the chip which deteriorates the cooling. It turns out to be a real lottery in which all depends on a particular specimen of the card. You have a chance of wining a 9700 card, but you may also win stripes and overheating or, worse than that, you can either win it all or win nothing at all.. What makes the card a bit more attractive is the price which is also not so low for a card of such performance index. With a great reserve, this card can be used in the not yet released games, since its functionality it compensated by low speed characteristics solvable through overclocking, if any :). The disconnection of 4 pipelines together with the HyperZ module, plus cutting the bit-rate down to 128 bit from 256 has resulted in serious performance losses, to which the benchmarks point. We intentionally accentuated the processor dependence during the tests since the card is positioned on the market as aimed at the mid-range sector. The main conclusion is that the most justified option would be installing an ATI 9500Pro on the processor of the AMD XP 2000+ line, since with such performance level the processor is no longer a bottleneck and won't restrict the video system. As for the implementation of the card, Gigabyte has got nothing to claim about, moreover, the package bundle really deserves a praise. What was really poor is the adapter cable to plug in TV-Out, but that is solvable through purchasing regular video extension cables. And one more advice - but the video cards at sales places where they provide a MoneyBack guarantee or have well-established guarantee servicing in order to have a good reason for having your card replaced in case stripes are seen on the screen.

Pros:

• Full-featured hardware support for DirectX 9.
• Random possibility to turn the card into a 9700 analog.
• Good frequency overclocking.
• Good package bundle.

Cons:

• Low 3D performance.
• Possible hardware flaws (on-screen waves in 2D, etc).
• Slightly buggy drivers.
• High price.